Daily Papers

Singing, as a common facial movement second only to talking, can be regarded as a universal language across ethnicities and cultures, plays an important role in emotional communication, art, and entertainment. However, it is often overlooked in the field of audio-driven facial animation due to the lack of singing head datasets and the domain gap between singing and talking in rhythm and amplitude. To this end, we collect a high-quality large-scale singing head dataset, SingingHead, which consists of more than 27 hours of synchronized singing video, 3D facial motion, singing audio, and background music from 76 individuals and 8 types of music. Along with the SingingHead dataset, we benchmark existing audio-driven 3D facial animation methods and 2D talking head methods on the singing task. Furthermore, we argue that 3D and 2D facial animation tasks can be solved together, and propose a unified singing head animation framework named UniSinger to achieve both singing audio-driven 3D singing head animation and 2D singing portrait video synthesis, which achieves competitive results on both 3D and 2D benchmarks. Extensive experiments demonstrate the significance of the proposed singing-specific dataset in promoting the development of singing head animation tasks, as well as the promising performance of our unified facial animation framework.

We propose a unified framework for Singing Voice Synthesis (SVS) and Conversion (SVC), addressing the limitations of existing approaches in cross-domain SVS/SVC, poor output musicality, and scarcity of singing data. Our framework enables control over multiple aspects, including language content based on lyrics, performance attributes based on a musical score, singing style and vocal techniques based on a selector, and voice identity based on a speech sample. The proposed zero-shot learning paradigm consists of one SVS model and two SVC models, utilizing pre-trained content embeddings and a diffusion-based generator. The proposed framework is also trained on mixed datasets comprising both singing and speech audio, allowing singing voice cloning based on speech reference. Experiments show substantial improvements in timbre similarity and musicality over state-of-the-art baselines, providing insights into other low-data music tasks such as instrumental style transfer. Examples can be found at: everyone-can-sing.github.io.

Developing a versatile deep neural network to model music audio is crucial in MIR. This task is challenging due to the intricate spectral variations inherent in music signals, which convey melody, harmonics, and timbres of diverse instruments. In this paper, we introduce Mel-RoFormer, a spectrogram-based model featuring two key designs: a novel Mel-band Projection module at the front-end to enhance the model's capability to capture informative features across multiple frequency bands, and interleaved RoPE Transformers to explicitly model the frequency and time dimensions as two separate sequences. We apply Mel-RoFormer to tackle two essential MIR tasks: vocal separation and vocal melody transcription, aimed at isolating singing voices from audio mixtures and transcribing their lead melodies, respectively. Despite their shared focus on singing signals, these tasks possess distinct optimization objectives. Instead of training a unified model, we adopt a two-step approach. Initially, we train a vocal separation model, which subsequently serves as a foundation model for fine-tuning for vocal melody transcription. Through extensive experiments conducted on benchmark datasets, we showcase that our models achieve state-of-the-art performance in both vocal separation and melody transcription tasks, underscoring the efficacy and versatility of Mel-RoFormer in modeling complex music audio signals.

Cinematic audio source separation (CASS) is a fairly new subtask of audio source separation. A typical setup of CASS is a three-stem problem, with the aim of separating the mixture into the dialogue stem (DX), music stem (MX), and effects stem (FX). In practice, however, several edge cases exist as some sound sources do not fit neatly in either of these three stems, necessitating the use of additional auxiliary stems in production. One very common edge case is the singing voice in film audio, which may belong in either the DX or MX, depending heavily on the cinematic context. In this work, we demonstrate a very straightforward extension of the dedicated-decoder Bandit and query-based single-decoder Banquet models to a four-stem problem, treating non-musical dialogue, instrumental music, singing voice, and effects as separate stems. Interestingly, the query-based Banquet model outperformed the dedicated-decoder Bandit model. We hypothesized that this is due to a better feature alignment at the bottleneck as enforced by the band-agnostic FiLM layer. Dataset and model implementation will be made available at https://github.com/kwatcharasupat/source-separation-landing.

Recently, denoising diffusion models have demonstrated remarkable performance among generative models in various domains. However, in the speech domain, the application of diffusion models for synthesizing time-varying audio faces limitations in terms of complexity and controllability, as speech synthesis requires very high-dimensional samples with long-term acoustic features. To alleviate the challenges posed by model complexity in singing voice synthesis, we propose HiddenSinger, a high-quality singing voice synthesis system using a neural audio codec and latent diffusion models. To ensure high-fidelity audio, we introduce an audio autoencoder that can encode audio into an audio codec as a compressed representation and reconstruct the high-fidelity audio from the low-dimensional compressed latent vector. Subsequently, we use the latent diffusion models to sample a latent representation from a musical score. In addition, our proposed model is extended to an unsupervised singing voice learning framework, HiddenSinger-U, to train the model using an unlabeled singing voice dataset. Experimental results demonstrate that our model outperforms previous models in terms of audio quality. Furthermore, the HiddenSinger-U can synthesize high-quality singing voices of speakers trained solely on unlabeled data.

This paper presents a benchmark for singing voice enhancement. The development of singing voice enhancement is limited by the lack of realistic evaluation data. To address this gap, this paper introduces SingVERSE, the first real-world benchmark for singing voice enhancement, covering diverse acoustic scenarios and providing paired, studio-quality clean references. Leveraging SingVERSE, we conduct a comprehensive evaluation of state-of-the-art models and uncover a consistent trade-off between perceptual quality and intelligibility. Finally, we show that training on in-domain singing data substantially improves enhancement performance without degrading speech capabilities, establishing a simple yet effective path forward. This work offers the community a foundational benchmark together with critical insights to guide future advances in this underexplored domain. Demopage: https://singverse.github.io

Singing voice synthesis and conversion have emerged as significant subdomains of voice generation, leading to much demands on prompt-conditioned generation. Unlike common voice data, generating a singing voice requires an understanding of various associated vocal and musical characteristics, such as the vocal tone of the singer or emotional expressions. However, existing open-source audio-text datasets for voice generation tend to capture only a very limited range of attributes, often missing musical characteristics of the audio. To fill this gap, we introduce S2Cap, an audio-text pair dataset with a diverse set of attributes. S2Cap consists of pairs of textual prompts and music audio samples with a wide range of vocal and musical attributes, including pitch, volume, tempo, mood, singer's gender and age, and musical genre and emotional expression. Utilizing S2Cap, we suggest an effective novel baseline algorithm for singing style captioning. Singing style captioning is a relative task to voice generation that generates text descriptions of vocal characteristics, which we first suggested. First, to mitigate the misalignment between the audio encoder and the text decoder, we present a novel mechanism called CRESCENDO, which utilizes positive-pair similarity learning to synchronize the embedding spaces of a pretrained audio encoder to get similar embeddings with a text encoder. We additionally supervise the model using the singer's voice, which is demixed by the accompaniment. This supervision allows the model to more accurately capture vocal characteristics, leading to improved singing style captions that better reflect the style of the singer. The dataset and the codes are available at https://github.com/HJ-Ok/S2cap.

Note-level Automatic Singing Voice Transcription (AST) converts singing recordings into note sequences, facilitating the automatic annotation of singing datasets for Singing Voice Synthesis (SVS) applications. Current AST methods, however, struggle with accuracy and robustness when used for practical annotation. This paper presents ROSVOT, the first robust AST model that serves SVS, incorporating a multi-scale framework that effectively captures coarse-grained note information and ensures fine-grained frame-level segmentation, coupled with an attention-based pitch decoder for reliable pitch prediction. We also established a comprehensive annotation-and-training pipeline for SVS to test the model in real-world settings. Experimental findings reveal that ROSVOT achieves state-of-the-art transcription accuracy with either clean or noisy inputs. Moreover, when trained on enlarged, automatically annotated datasets, the SVS model outperforms its baseline, affirming the capability for practical application. Audio samples are available at https://rosvot.github.io.

Text-to-song (TTSong) is a music generation task that synthesizes accompanied singing voices. Current TTSong methods, inherited from singing voice synthesis (SVS), require melody-related information that can sometimes be impractical, such as music scores or MIDI sequences. We present MelodyLM, the first TTSong model that generates high-quality song pieces with fully text-controlled melodies, achieving minimal user requirements and maximum control flexibility. MelodyLM explicitly models MIDI as the intermediate melody-related feature and sequentially generates vocal tracks in a language model manner, conditioned on textual and vocal prompts. The accompaniment music is subsequently synthesized by a latent diffusion model with hybrid conditioning for temporal alignment. With minimal requirements, users only need to input lyrics and a reference voice to synthesize a song sample. For full control, just input textual prompts or even directly input MIDI. Experimental results indicate that MelodyLM achieves superior performance in terms of both objective and subjective metrics. Audio samples are available at https://melodylm666.github.io.

Singing voice synthesis (SVS) systems are built to synthesize high-quality and expressive singing voice, in which the acoustic model generates the acoustic features (e.g., mel-spectrogram) given a music score. Previous singing acoustic models adopt a simple loss (e.g., L1 and L2) or generative adversarial network (GAN) to reconstruct the acoustic features, while they suffer from over-smoothing and unstable training issues respectively, which hinder the naturalness of synthesized singing. In this work, we propose DiffSinger, an acoustic model for SVS based on the diffusion probabilistic model. DiffSinger is a parameterized Markov chain that iteratively converts the noise into mel-spectrogram conditioned on the music score. By implicitly optimizing variational bound, DiffSinger can be stably trained and generate realistic outputs. To further improve the voice quality and speed up inference, we introduce a shallow diffusion mechanism to make better use of the prior knowledge learned by the simple loss. Specifically, DiffSinger starts generation at a shallow step smaller than the total number of diffusion steps, according to the intersection of the diffusion trajectories of the ground-truth mel-spectrogram and the one predicted by a simple mel-spectrogram decoder. Besides, we propose boundary prediction methods to locate the intersection and determine the shallow step adaptively. The evaluations conducted on a Chinese singing dataset demonstrate that DiffSinger outperforms state-of-the-art SVS work. Extensional experiments also prove the generalization of our methods on text-to-speech task (DiffSpeech). Audio samples: https://diffsinger.github.io. Codes: https://github.com/MoonInTheRiver/DiffSinger. The old title of this work: "Diffsinger: Diffusion acoustic model for singing voice synthesis".

The diffusion-based Singing Voice Conversion (SVC) methods have achieved remarkable performances, producing natural audios with high similarity to the target timbre. However, the iterative sampling process results in slow inference speed, and acceleration thus becomes crucial. In this paper, we propose CoMoSVC, a consistency model-based SVC method, which aims to achieve both high-quality generation and high-speed sampling. A diffusion-based teacher model is first specially designed for SVC, and a student model is further distilled under self-consistency properties to achieve one-step sampling. Experiments on a single NVIDIA GTX4090 GPU reveal that although CoMoSVC has a significantly faster inference speed than the state-of-the-art (SOTA) diffusion-based SVC system, it still achieves comparable or superior conversion performance based on both subjective and objective metrics. Audio samples and codes are available at https://comosvc.github.io/.

Singing voice synthesis has made remarkable progress in generating natural and high-quality voices. However, existing methods rarely provide precise control over vocal techniques such as intensity, mixed voice, falsetto, bubble, and breathy tones, thus limiting the expressive potential of synthetic voices. We introduce TechSinger, an advanced system for controllable singing voice synthesis that supports five languages and seven vocal techniques. TechSinger leverages a flow-matching-based generative model to produce singing voices with enhanced expressive control over various techniques. To enhance the diversity of training data, we develop a technique detection model that automatically annotates datasets with phoneme-level technique labels. Additionally, our prompt-based technique prediction model enables users to specify desired vocal attributes through natural language, offering fine-grained control over the synthesized singing. Experimental results demonstrate that TechSinger significantly enhances the expressiveness and realism of synthetic singing voices, outperforming existing methods in terms of audio quality and technique-specific control. Audio samples can be found at https://tech-singer.github.io.

Although Singing Voice Synthesis (SVS) has made great strides with Text-to-Speech (TTS) techniques, multilingual singing voice modeling remains relatively unexplored. This paper presents BiSinger, a bilingual pop SVS system for English and Chinese Mandarin. Current systems require separate models per language and cannot accurately represent both Chinese and English, hindering code-switch SVS. To address this gap, we design a shared representation between Chinese and English singing voices, achieved by using the CMU dictionary with mapping rules. We fuse monolingual singing datasets with open-source singing voice conversion techniques to generate bilingual singing voices while also exploring the potential use of bilingual speech data. Experiments affirm that our language-independent representation and incorporation of related datasets enable a single model with enhanced performance in English and code-switch SVS while maintaining Chinese song performance. Audio samples are available at https://bisinger-svs.github.io.

Recent singing-voice-synthesis (SVS) methods have achieved remarkable audio quality and naturalness, yet they lack the capability to control the style attributes of the synthesized singing explicitly. We propose Prompt-Singer, the first SVS method that enables attribute controlling on singer gender, vocal range and volume with natural language. We adopt a model architecture based on a decoder-only transformer with a multi-scale hierarchy, and design a range-melody decoupled pitch representation that enables text-conditioned vocal range control while keeping melodic accuracy. Furthermore, we explore various experiment settings, including different types of text representations, text encoder fine-tuning, and introducing speech data to alleviate data scarcity, aiming to facilitate further research. Experiments show that our model achieves favorable controlling ability and audio quality. Audio samples are available at http://prompt-singer.github.io .

Denoising diffusion probabilistic models (DDPMs) have shown promising performance for speech synthesis. However, a large number of iterative steps are required to achieve high sample quality, which restricts the inference speed. Maintaining sample quality while increasing sampling speed has become a challenging task. In this paper, we propose a "Co"nsistency "Mo"del-based "Speech" synthesis method, CoMoSpeech, which achieve speech synthesis through a single diffusion sampling step while achieving high audio quality. The consistency constraint is applied to distill a consistency model from a well-designed diffusion-based teacher model, which ultimately yields superior performances in the distilled CoMoSpeech. Our experiments show that by generating audio recordings by a single sampling step, the CoMoSpeech achieves an inference speed more than 150 times faster than real-time on a single NVIDIA A100 GPU, which is comparable to FastSpeech2, making diffusion-sampling based speech synthesis truly practical. Meanwhile, objective and subjective evaluations on text-to-speech and singing voice synthesis show that the proposed teacher models yield the best audio quality, and the one-step sampling based CoMoSpeech achieves the best inference speed with better or comparable audio quality to other conventional multi-step diffusion model baselines. Audio samples are available at https://comospeech.github.io/.

Customizable multilingual zero-shot singing voice synthesis (SVS) has various potential applications in music composition and short video dubbing. However, existing SVS models overly depend on phoneme and note boundary annotations, limiting their robustness in zero-shot scenarios and producing poor transitions between phonemes and notes. Moreover, they also lack effective multi-level style control via diverse prompts. To overcome these challenges, we introduce TCSinger 2, a multi-task multilingual zero-shot SVS model with style transfer and style control based on various prompts. TCSinger 2 mainly includes three key modules: 1) Blurred Boundary Content (BBC) Encoder, predicts duration, extends content embedding, and applies masking to the boundaries to enable smooth transitions. 2) Custom Audio Encoder, uses contrastive learning to extract aligned representations from singing, speech, and textual prompts. 3) Flow-based Custom Transformer, leverages Cus-MOE, with F0 supervision, enhancing both the synthesis quality and style modeling of the generated singing voice. Experimental results show that TCSinger 2 outperforms baseline models in both subjective and objective metrics across multiple related tasks. Singing voice samples are available at https://aaronz345.github.io/TCSinger2Demo/.

We are interested in a novel task, singing voice beautifying (SVB). Given the singing voice of an amateur singer, SVB aims to improve the intonation and vocal tone of the voice, while keeping the content and vocal timbre. Current automatic pitch correction techniques are immature, and most of them are restricted to intonation but ignore the overall aesthetic quality. Hence, we introduce Neural Singing Voice Beautifier (NSVB), the first generative model to solve the SVB task, which adopts a conditional variational autoencoder as the backbone and learns the latent representations of vocal tone. In NSVB, we propose a novel time-warping approach for pitch correction: Shape-Aware Dynamic Time Warping (SADTW), which ameliorates the robustness of existing time-warping approaches, to synchronize the amateur recording with the template pitch curve. Furthermore, we propose a latent-mapping algorithm in the latent space to convert the amateur vocal tone to the professional one. To achieve this, we also propose a new dataset containing parallel singing recordings of both amateur and professional versions. Extensive experiments on both Chinese and English songs demonstrate the effectiveness of our methods in terms of both objective and subjective metrics. Audio samples are available at~https://neuralsvb.github.io. Codes: https://github.com/MoonInTheRiver/NeuralSVB.

Controllable human voice generation, particularly for expressive domains like singing, remains a significant challenge. This paper introduces Vevo2, a unified framework for controllable speech and singing voice generation. To tackle issues like the scarcity of annotated singing data and to enable flexible controllability, Vevo2 introduces two audio tokenizers: (1) a music-notation-free prosody tokenizer that captures prosody and melody from speech, singing, and even instrumental sounds, and (2) a low-frame-rate (12.5 Hz) content-style tokenizer that encodes linguistic content, prosody, and style for both speech and singing, while enabling timbre disentanglement. Vevo2 consists of an auto-regressive (AR) content-style modeling stage, which aims to enable controllability over text, prosody, and style, as well as a flow-matching acoustic modeling stage that allows for timbre control. Particularly, during pre-training of the AR model, we propose both explicit and implicit prosody learning strategies to bridge speech and singing voice. Moreover, to further enhance the AR model's ability to follow text and prosody, we design a multi-objective post-training task that integrates both intelligibility and prosody similarity alignment. Experimental results show that the unified modeling in Vevo2 brings mutual benefits to both speech and singing voice generation. Additionally, Vevo2's effectiveness across a wide range of synthesis, conversion, and editing tasks for both speech and singing further demonstrates its strong generalization ability and versatility. Audio samples are are available at https://versasinger.github.io/.

Recent breakthroughs in singing voice synthesis (SVS) have heightened the demand for high-quality annotated datasets, yet manual annotation remains prohibitively labor-intensive and resource-intensive. Existing automatic singing annotation (ASA) methods, however, primarily tackle isolated aspects of the annotation pipeline. To address this fundamental challenge, we present STARS, which is, to our knowledge, the first unified framework that simultaneously addresses singing transcription, alignment, and refined style annotation. Our framework delivers comprehensive multi-level annotations encompassing: (1) precise phoneme-audio alignment, (2) robust note transcription and temporal localization, (3) expressive vocal technique identification, and (4) global stylistic characterization including emotion and pace. The proposed architecture employs hierarchical acoustic feature processing across frame, word, phoneme, note, and sentence levels. The novel non-autoregressive local acoustic encoders enable structured hierarchical representation learning. Experimental validation confirms the framework's superior performance across multiple evaluation dimensions compared to existing annotation approaches. Furthermore, applications in SVS training demonstrate that models utilizing STARS-annotated data achieve significantly enhanced perceptual naturalness and precise style control. This work not only overcomes critical scalability challenges in the creation of singing datasets but also pioneers new methodologies for controllable singing voice synthesis. Audio samples are available at https://gwx314.github.io/stars-demo/.

Separation of multiple singing voices into each voice is a rarely studied area in music source separation research. The absence of a benchmark dataset has hindered its progress. In this paper, we present an evaluation dataset and provide baseline studies for multiple singing voices separation. First, we introduce MedleyVox, an evaluation dataset for multiple singing voices separation. We specify the problem definition in this dataset by categorizing it into i) unison, ii) duet, iii) main vs. rest, and iv) N-singing separation. Second, to overcome the absence of existing multi-singing datasets for a training purpose, we present a strategy for construction of multiple singing mixtures using various single-singing datasets. Third, we propose the improved super-resolution network (iSRNet), which greatly enhances initial estimates of separation networks. Jointly trained with the Conv-TasNet and the multi-singing mixture construction strategy, the proposed iSRNet achieved comparable performance to ideal time-frequency masks on duet and unison subsets of MedleyVox. Audio samples, the dataset, and codes are available on our website (https://github.com/jeonchangbin49/MedleyVox).

This research presents Muskits-ESPnet, a versatile toolkit that introduces new paradigms to Singing Voice Synthesis (SVS) through the application of pretrained audio models in both continuous and discrete approaches. Specifically, we explore discrete representations derived from SSL models and audio codecs and offer significant advantages in versatility and intelligence, supporting multi-format inputs and adaptable data processing workflows for various SVS models. The toolkit features automatic music score error detection and correction, as well as a perception auto-evaluation module to imitate human subjective evaluating scores. Muskits-ESPnet is available at https://github.com/espnet/espnet.

This paper introduces Opencpop, a publicly available high-quality Mandarin singing corpus designed for singing voice synthesis (SVS). The corpus consists of 100 popular Mandarin songs performed by a female professional singer. Audio files are recorded with studio quality at a sampling rate of 44,100 Hz and the corresponding lyrics and musical scores are provided. All singing recordings have been phonetically annotated with phoneme boundaries and syllable (note) boundaries. To demonstrate the reliability of the released data and to provide a baseline for future research, we built baseline deep neural network-based SVS models and evaluated them with both objective metrics and subjective mean opinion score (MOS) measure. Experimental results show that the best SVS model trained on our database achieves 3.70 MOS, indicating the reliability of the provided corpus. Opencpop is released to the open-source community WeNet, and the corpus, as well as synthesized demos, can be found on the project homepage.

Any-to-any singing voice conversion is confronted with a significant challenge of ``timbre leakage'' issue caused by inadequate disentanglement between the content and the speaker timbre. To address this issue, this study introduces a novel neural concatenative singing voice conversion (NeuCoSVC) framework. The NeuCoSVC framework comprises a self-supervised learning (SSL) representation extractor, a neural harmonic signal generator, and a waveform synthesizer. Specifically, the SSL extractor condenses the audio into a sequence of fixed-dimensional SSL features. The harmonic signal generator produces both raw and filtered harmonic signals as the pitch information by leveraging a linear time-varying (LTV) filter. Finally, the audio generator reconstructs the audio waveform based on the SSL features, as well as the harmonic signals and the loudness information. During inference, the system performs voice conversion by substituting source SSL features with their nearest counterparts from a matching pool, which comprises SSL representations extracted from the target audio, while the raw harmonic signals and the loudness are extracted from the source audio and are kept unchanged. Since the utilized SSL features in the conversion stage are directly from the target audio, the proposed framework has great potential to address the ``timbre leakage'' issue caused by previous disentanglement-based approaches. Experimental results confirm that the proposed system delivers much better performance than the speaker embedding approach (disentanglement-based) in the context of one-shot SVC across intra-language, cross-language, and cross-domain evaluations.

Scaling text-to-speech (TTS) to large-scale, multi-speaker, and in-the-wild datasets is important to capture the diversity in human speech such as speaker identities, prosodies, and styles (e.g., singing). Current large TTS systems usually quantize speech into discrete tokens and use language models to generate these tokens one by one, which suffer from unstable prosody, word skipping/repeating issue, and poor voice quality. In this paper, we develop NaturalSpeech 2, a TTS system that leverages a neural audio codec with residual vector quantizers to get the quantized latent vectors and uses a diffusion model to generate these latent vectors conditioned on text input. To enhance the zero-shot capability that is important to achieve diverse speech synthesis, we design a speech prompting mechanism to facilitate in-context learning in the diffusion model and the duration/pitch predictor. We scale NaturalSpeech 2 to large-scale datasets with 44K hours of speech and singing data and evaluate its voice quality on unseen speakers. NaturalSpeech 2 outperforms previous TTS systems by a large margin in terms of prosody/timbre similarity, robustness, and voice quality in a zero-shot setting, and performs novel zero-shot singing synthesis with only a speech prompt. Audio samples are available at https://speechresearch.github.io/naturalspeech2.

Models for audio source separation usually operate on the magnitude spectrum, which ignores phase information and makes separation performance dependant on hyper-parameters for the spectral front-end. Therefore, we investigate end-to-end source separation in the time-domain, which allows modelling phase information and avoids fixed spectral transformations. Due to high sampling rates for audio, employing a long temporal input context on the sample level is difficult, but required for high quality separation results because of long-range temporal correlations. In this context, we propose the Wave-U-Net, an adaptation of the U-Net to the one-dimensional time domain, which repeatedly resamples feature maps to compute and combine features at different time scales. We introduce further architectural improvements, including an output layer that enforces source additivity, an upsampling technique and a context-aware prediction framework to reduce output artifacts. Experiments for singing voice separation indicate that our architecture yields a performance comparable to a state-of-the-art spectrogram-based U-Net architecture, given the same data. Finally, we reveal a problem with outliers in the currently used SDR evaluation metrics and suggest reporting rank-based statistics to alleviate this problem.

The advent of Large Models marks a new era in machine learning, significantly outperforming smaller models by leveraging vast datasets to capture and synthesize complex patterns. Despite these advancements, the exploration into scaling, especially in the audio generation domain, remains limited, with previous efforts didn't extend into the high-fidelity (HiFi) 44.1kHz domain and suffering from both spectral discontinuities and blurriness in the high-frequency domain, alongside a lack of robustness against out-of-domain data. These limitations restrict the applicability of models to diverse use cases, including music and singing generation. Our work introduces Enhanced Various Audio Generation via Scalable Generative Adversarial Networks (EVA-GAN), yields significant improvements over previous state-of-the-art in spectral and high-frequency reconstruction and robustness in out-of-domain data performance, enabling the generation of HiFi audios by employing an extensive dataset of 36,000 hours of 44.1kHz audio, a context-aware module, a Human-In-The-Loop artifact measurement toolkit, and expands the model to approximately 200 million parameters. Demonstrations of our work are available at https://double-blind-eva-gan.cc.

Real-time speech interaction, serving as a fundamental interface for human-machine collaboration, holds immense potential. However, current open-source models face limitations such as high costs in voice data collection, weakness in dynamic control, and limited intelligence. To address these challenges, this paper introduces Step-Audio, the first production-ready open-source solution. Key contributions include: 1) a 130B-parameter unified speech-text multi-modal model that achieves unified understanding and generation, with the Step-Audio-Chat version open-sourced; 2) a generative speech data engine that establishes an affordable voice cloning framework and produces the open-sourced lightweight Step-Audio-TTS-3B model through distillation; 3) an instruction-driven fine control system enabling dynamic adjustments across dialects, emotions, singing, and RAP; 4) an enhanced cognitive architecture augmented with tool calling and role-playing abilities to manage complex tasks effectively. Based on our new StepEval-Audio-360 evaluation benchmark, Step-Audio achieves state-of-the-art performance in human evaluations, especially in terms of instruction following. On open-source benchmarks like LLaMA Question, shows 9.3% average performance improvement, demonstrating our commitment to advancing the development of open-source multi-modal language technologies. Our code and models are available at https://github.com/stepfun-ai/Step-Audio.

Neural vocoders and codecs reconstruct waveforms from acoustic representations, which directly impact the audio quality. Among existing methods, upsampling-based time-domain models are superior in both inference speed and synthesis quality, achieving state-of-the-art performance. Still, despite their success in producing perceptually natural sound, their synthesis fidelity remains limited due to the aliasing artifacts brought by the inadequately designed model architectures. In particular, the unconstrained nonlinear activation generates an infinite number of harmonics that exceed the Nyquist frequency, resulting in ``folded-back'' aliasing artifacts. The widely used upsampling layer, ConvTranspose, copies the mirrored low-frequency parts to fill the empty high-frequency region, resulting in ``mirrored'' aliasing artifacts. Meanwhile, the combination of its inherent periodicity and the mirrored DC bias also brings ``tonal artifact,'' resulting in constant-frequency ringing. This paper aims to solve these issues from a signal processing perspective. Specifically, we apply oversampling and anti-derivative anti-aliasing to the activation function to obtain its anti-aliased form, and replace the problematic ConvTranspose layer with resampling to avoid the ``tonal artifact'' and eliminate aliased components. Based on our proposed anti-aliased modules, we introduce Pupu-Vocoder and Pupu-Codec, and release high-quality pre-trained checkpoints to facilitate audio generation research. We build a test signal benchmark to illustrate the effectiveness of the anti-aliased modules, and conduct experiments on speech, singing voice, music, and audio to validate our proposed models. Experimental results confirm that our lightweight Pupu-Vocoder and Pupu-Codec models can easily outperform existing systems on singing voice, music, and audio, while achieving comparable performance on speech.

Traditional Blind Source Separation Evaluation (BSS-Eval) metrics were originally designed to evaluate linear audio source separation models based on methods such as time-frequency masking. However, recent generative models may introduce nonlinear relationships between the separated and reference signals, limiting the reliability of these metrics for objective evaluation. To address this issue, we conduct a Degradation Category Rating listening test and analyze correlations between the obtained degradation mean opinion scores (DMOS) and a set of objective audio quality metrics for the task of singing voice separation. We evaluate three state-of-the-art discriminative models and two new competitive generative models. For both discriminative and generative models, intrusive embedding-based metrics show higher correlations with DMOS than conventional intrusive metrics such as BSS-Eval. For discriminative models, the highest correlation is achieved by the MSE computed on Music2Latent embeddings. When it comes to the evaluation of generative models, the strongest correlations are evident for the multi-resolution STFT loss and the MSE calculated on MERT-L12 embeddings, with the latter also providing the most balanced correlation across both model types. Our results highlight the limitations of BSS-Eval metrics for evaluating generative singing voice separation models and emphasize the need for careful selection and validation of alternative evaluation metrics for the task of singing voice separation.

Audio-visual video segmentation~(AVVS) aims to generate pixel-level maps of sound-producing objects within image frames and ensure the maps faithfully adhere to the given audio, such as identifying and segmenting a singing person in a video. However, existing methods exhibit two limitations: 1) they address video temporal features and audio-visual interactive features separately, disregarding the inherent spatial-temporal dependence of combined audio and video, and 2) they inadequately introduce audio constraints and object-level information during the decoding stage, resulting in segmentation outcomes that fail to comply with audio directives. To tackle these issues, we propose a decoupled audio-video transformer that combines audio and video features from their respective temporal and spatial dimensions, capturing their combined dependence. To optimize memory consumption, we design a block, which, when stacked, enables capturing audio-visual fine-grained combinatorial-dependence in a memory-efficient manner. Additionally, we introduce audio-constrained queries during the decoding phase. These queries contain rich object-level information, ensuring the decoded mask adheres to the sounds. Experimental results confirm our approach's effectiveness, with our framework achieving a new SOTA performance on all three datasets using two backbones. The code is available at https://github.com/aspirinone/CATR.github.io

Language models (LMs) have demonstrated the capability to handle a variety of generative tasks. This paper presents the UniAudio system, which, unlike prior task-specific approaches, leverages LMs techniques to generate multiple types of audio (including speech, sounds, music, and singing) with given input conditions. UniAudio 1) first tokenizes all types of target audio along with other condition modalities, 2) concatenates source-target pair as a single sequence, and 3) performs next-token prediction using LMs. Also, a multi-scale Transformer model is proposed to handle the overly long sequences caused by the residual vector quantization based neural codec in tokenization. Training of UniAudio is scaled up to 165K hours of audio and 1B parameters, based on all generative tasks, aiming to obtain sufficient prior knowledge not only in the intrinsic properties of audio but also the inter-relationship between audio and other modalities. Therefore, the trained UniAudio model has the potential to become a foundation model for universal audio generation: it shows strong capability in all trained tasks and can seamlessly support new audio generation tasks after simple fine-tuning. Experiments demonstrate that UniAudio achieves state-of-the-art or at least competitive results on most of the 11 tasks. Demo and code are released at https://github.com/yangdongchao/UniAudio

Zero-shot singing voice synthesis (SVS) with style transfer and style control aims to generate high-quality singing voices with unseen timbres and styles (including singing method, emotion, rhythm, technique, and pronunciation) from audio and text prompts. However, the multifaceted nature of singing styles poses a significant challenge for effective modeling, transfer, and control. Furthermore, current SVS models often fail to generate singing voices rich in stylistic nuances for unseen singers. To address these challenges, we introduce TCSinger, the first zero-shot SVS model for style transfer across cross-lingual speech and singing styles, along with multi-level style control. Specifically, TCSinger proposes three primary modules: 1) the clustering style encoder employs a clustering vector quantization model to stably condense style information into a compact latent space; 2) the Style and Duration Language Model (S\&D-LM) concurrently predicts style information and phoneme duration, which benefits both; 3) the style adaptive decoder uses a novel mel-style adaptive normalization method to generate singing voices with enhanced details. Experimental results show that TCSinger outperforms all baseline models in synthesis quality, singer similarity, and style controllability across various tasks, including zero-shot style transfer, multi-level style control, cross-lingual style transfer, and speech-to-singing style transfer. Singing voice samples can be accessed at https://tcsinger.github.io/.

Zero-shot singing voice conversion (SVC) transforms a source singer's timbre to an unseen target speaker's voice while preserving melodic content without fine-tuning. Existing methods model speaker timbre and vocal content separately, losing essential acoustic information that degrades output quality while requiring significant computational resources. To overcome these limitations, we propose HQ-SVC, an efficient framework for high-quality zero-shot SVC. HQ-SVC first extracts jointly content and speaker features using a decoupled codec. It then enhances fidelity through pitch and volume modeling, preserving critical acoustic information typically lost in separate modeling approaches, and progressively refines outputs via differentiable signal processing and diffusion techniques. Evaluations confirm HQ-SVC significantly outperforms state-of-the-art zero-shot SVC methods in conversion quality and efficiency. Beyond voice conversion, HQ-SVC achieves superior voice naturalness compared to specialized audio super-resolution methods while natively supporting voice super-resolution tasks.

Style transfer for out-of-domain (OOD) singing voice synthesis (SVS) focuses on generating high-quality singing voices with unseen styles (such as timbre, emotion, pronunciation, and articulation skills) derived from reference singing voice samples. However, the endeavor to model the intricate nuances of singing voice styles is an arduous task, as singing voices possess a remarkable degree of expressiveness. Moreover, existing SVS methods encounter a decline in the quality of synthesized singing voices in OOD scenarios, as they rest upon the assumption that the target vocal attributes are discernible during the training phase. To overcome these challenges, we propose StyleSinger, the first singing voice synthesis model for zero-shot style transfer of out-of-domain reference singing voice samples. StyleSinger incorporates two critical approaches for enhanced effectiveness: 1) the Residual Style Adaptor (RSA) which employs a residual quantization module to capture diverse style characteristics in singing voices, and 2) the Uncertainty Modeling Layer Normalization (UMLN) to perturb the style attributes within the content representation during the training phase and thus improve the model generalization. Our extensive evaluations in zero-shot style transfer undeniably establish that StyleSinger outperforms baseline models in both audio quality and similarity to the reference singing voice samples. Access to singing voice samples can be found at https://stylesinger.github.io/.

Humans rely on multisensory integration to perceive spatial environments, where auditory cues enable sound source localization in three-dimensional space. Despite the critical role of spatial audio in immersive technologies such as VR/AR, most existing multimodal datasets provide only monaural audio, which limits the development of spatial audio generation and understanding. To address these challenges, we introduce MRSAudio, a large-scale multimodal spatial audio dataset designed to advance research in spatial audio understanding and generation. MRSAudio spans four distinct components: MRSLife, MRSSpeech, MRSMusic, and MRSSing, covering diverse real-world scenarios. The dataset includes synchronized binaural and ambisonic audio, exocentric and egocentric video, motion trajectories, and fine-grained annotations such as transcripts, phoneme boundaries, lyrics, scores, and prompts. To demonstrate the utility and versatility of MRSAudio, we establish five foundational tasks: audio spatialization, and spatial text to speech, spatial singing voice synthesis, spatial music generation and sound event localization and detection. Results show that MRSAudio enables high-quality spatial modeling and supports a broad range of spatial audio research. Demos and dataset access are available at https://mrsaudio.github.io.

Singing voice conversion (SVC) is a technique to enable an arbitrary singer to sing an arbitrary song. To achieve that, it is important to obtain speaker-agnostic representations from source audio, which is a challenging task. A common solution is to extract content-based features (e.g., PPGs) from a pretrained acoustic model. However, the choices for acoustic models are vast and varied. It is yet to be explored what characteristics of content features from different acoustic models are, and whether integrating multiple content features can help each other. Motivated by that, this study investigates three distinct content features, sourcing from WeNet, Whisper, and ContentVec, respectively. We explore their complementary roles in intelligibility, prosody, and conversion similarity for SVC. By integrating the multiple content features with a diffusion-based SVC model, our SVC system achieves superior conversion performance on both objective and subjective evaluation in comparison to a single source of content features. Our demo page and code can be available https://www.zhangxueyao.com/data/MultipleContentsSVC/index.html.

Deep neural networks have become an indispensable technique for audio source separation (ASS). It was recently reported that a variant of CNN architecture called MMDenseNet was successfully employed to solve the ASS problem of estimating source amplitudes, and state-of-the-art results were obtained for DSD100 dataset. To further enhance MMDenseNet, here we propose a novel architecture that integrates long short-term memory (LSTM) in multiple scales with skip connections to efficiently model long-term structures within an audio context. The experimental results show that the proposed method outperforms MMDenseNet, LSTM and a blend of the two networks. The number of parameters and processing time of the proposed model are significantly less than those for simple blending. Furthermore, the proposed method yields better results than those obtained using ideal binary masks for a singing voice separation task.

In this work, we tackle the challenge of enhancing the realism and expressiveness in talking head video generation by focusing on the dynamic and nuanced relationship between audio cues and facial movements. We identify the limitations of traditional techniques that often fail to capture the full spectrum of human expressions and the uniqueness of individual facial styles. To address these issues, we propose EMO, a novel framework that utilizes a direct audio-to-video synthesis approach, bypassing the need for intermediate 3D models or facial landmarks. Our method ensures seamless frame transitions and consistent identity preservation throughout the video, resulting in highly expressive and lifelike animations. Experimental results demonsrate that EMO is able to produce not only convincing speaking videos but also singing videos in various styles, significantly outperforming existing state-of-the-art methodologies in terms of expressiveness and realism.

We propose Kling-Foley, a large-scale multimodal Video-to-Audio generation model that synthesizes high-quality audio synchronized with video content. In Kling-Foley, we introduce multimodal diffusion transformers to model the interactions between video, audio, and text modalities, and combine it with a visual semantic representation module and an audio-visual synchronization module to enhance alignment capabilities. Specifically, these modules align video conditions with latent audio elements at the frame level, thereby improving semantic alignment and audio-visual synchronization. Together with text conditions, this integrated approach enables precise generation of video-matching sound effects. In addition, we propose a universal latent audio codec that can achieve high-quality modeling in various scenarios such as sound effects, speech, singing, and music. We employ a stereo rendering method that imbues synthesized audio with a spatial presence. At the same time, in order to make up for the incomplete types and annotations of the open-source benchmark, we also open-source an industrial-level benchmark Kling-Audio-Eval. Our experiments show that Kling-Foley trained with the flow matching objective achieves new audio-visual SOTA performance among public models in terms of distribution matching, semantic alignment, temporal alignment and audio quality.

The scarcity of high-quality and multi-task singing datasets significantly hinders the development of diverse controllable and personalized singing tasks, as existing singing datasets suffer from low quality, limited diversity of languages and singers, absence of multi-technique information and realistic music scores, and poor task suitability. To tackle these problems, we present GTSinger, a large Global, multi-Technique, free-to-use, high-quality singing corpus with realistic music scores, designed for all singing tasks, along with its benchmarks. Particularly, (1) we collect 80.59 hours of high-quality singing voices, forming the largest recorded singing dataset; (2) 20 professional singers across nine widely spoken languages offer diverse timbres and styles; (3) we provide controlled comparison and phoneme-level annotations of six commonly used singing techniques, helping technique modeling and control; (4) GTSinger offers realistic music scores, assisting real-world musical composition; (5) singing voices are accompanied by manual phoneme-to-audio alignments, global style labels, and 16.16 hours of paired speech for various singing tasks. Moreover, to facilitate the use of GTSinger, we conduct four benchmark experiments: technique-controllable singing voice synthesis, technique recognition, style transfer, and speech-to-singing conversion. The corpus and demos can be found at http://gtsinger.github.io. We provide the dataset and the code for processing data and conducting benchmarks at https://huggingface.co/datasets/GTSinger/GTSinger and https://github.com/GTSinger/GTSinger.

In this work, we introduce VERSA, a unified and standardized evaluation toolkit designed for various speech, audio, and music signals. The toolkit features a Pythonic interface with flexible configuration and dependency control, making it user-friendly and efficient. With full installation, VERSA offers 63 metrics with 711 metric variations based on different configurations. These metrics encompass evaluations utilizing diverse external resources, including matching and non-matching reference audio, text transcriptions, and text captions. As a lightweight yet comprehensive toolkit, VERSA is versatile to support the evaluation of a wide range of downstream scenarios. To demonstrate its capabilities, this work highlights example use cases for VERSA, including audio coding, speech synthesis, speech enhancement, singing synthesis, and music generation. The toolkit is available at https://github.com/shinjiwlab/versa.

In singing voice synthesis (SVS), generating singing voices from musical scores faces challenges due to limited data availability, a constraint less common in text-to-speech (TTS). This study proposes a new approach to address this data scarcity. We utilize an existing singing voice synthesizer for data augmentation and apply precise manual tuning to reduce unnatural voice synthesis. Our development of two extensive singing voice corpora, ACE-Opencpop and KiSing-v2, facilitates large-scale, multi-singer voice synthesis. Utilizing pre-trained models derived from these corpora, we achieve notable improvements in voice quality, evident in both in-domain and out-of-domain scenarios. The corpora, pre-trained models, and their related training recipes are publicly available at Muskits-ESPnet (https://github.com/espnet/espnet).

Singing voice generation progresses rapidly, yet evaluating singing quality remains a critical challenge. Human subjective assessment, typically in the form of listening tests, is costly and time consuming, while existing objective metrics capture only limited perceptual aspects. In this work, we introduce SingMOS-Pro, a dataset for automatic singing quality assessment. Building on our preview version SingMOS, which provides only overall ratings, SingMOS-Pro expands annotations of the additional part to include lyrics, melody, and overall quality, offering broader coverage and greater diversity. The dataset contains 7,981 singing clips generated by 41 models across 12 datasets, spanning from early systems to recent advances. Each clip receives at least five ratings from professional annotators, ensuring reliability and consistency. Furthermore, we explore how to effectively utilize MOS data annotated under different standards and benchmark several widely used evaluation methods from related tasks on SingMOS-Pro, establishing strong baselines and practical references for future research. The dataset can be accessed at https://huggingface.co/datasets/TangRain/SingMOS-Pro.

Automated singing assessment is crucial for education and entertainment. However, existing systems face two fundamental limitations: reliance on reference tracks, which stifles creative expression, and the simplification of complex performances into non-diagnostic scores based solely on pitch and rhythm. We advocate for a shift from discriminative to descriptive evaluation, creating a complete ecosystem for reference-free, multi-dimensional assessment. First, we introduce Sing-MD, a large-scale dataset annotated by experts across four dimensions: breath control, timbre quality, emotional expression, and vocal technique. Our analysis reveals significant annotation inconsistencies among experts, challenging the validity of traditional accuracy-based metrics. Second, addressing the memory limitations of Multimodal Large Language Models (MLLMs) in analyzing full-length songs, we propose VocalVerse. This efficient hybrid architecture leverages a lightweight acoustic encoder to model global performance features and long-term dependencies. Third, to address automated metric shortcomings, we establish the H-TPR (Human-in-the-loop Tiered Perceptual Ranking) benchmark, which evaluates a model's ability to generate perceptually valid rankings rather than predicting noisy ground-truth scores.

This paper proposes singing voice synthesis (SVS) based on frame-level sequence-to-sequence models considering vocal timing deviation. In SVS, it is essential to synchronize the timing of singing with temporal structures represented by scores, taking into account that there are differences between actual vocal timing and note start timing. In many SVS systems including our previous work, phoneme-level score features are converted into frame-level ones on the basis of phoneme boundaries obtained by external aligners to take into account vocal timing deviations. Therefore, the sound quality is affected by the aligner accuracy in this system. To alleviate this problem, we introduce an attention mechanism with frame-level features. In the proposed system, the attention mechanism absorbs alignment errors in phoneme boundaries. Additionally, we evaluate the system with pseudo-phoneme-boundaries defined by heuristic rules based on musical scores when there is no aligner. The experimental results show the effectiveness of the proposed system.

Singing is one of the most cherished forms of human entertainment. However, creating a beautiful song requires an accompaniment that complements the vocals and aligns well with the song instruments and genre. With advancements in deep learning, previous research has focused on generating suitable accompaniments but often lacks precise alignment with the desired instrumentation and genre. To address this, we propose a straightforward method that enables control over the accompaniment through text prompts, allowing the generation of music that complements the vocals and aligns with the song instrumental and genre requirements. Through extensive experiments, we successfully generate 10-second accompaniments using vocal input and text control.

Pitch estimation (PE) in monophonic audio is crucial for MIDI transcription and singing voice conversion (SVC), but existing methods suffer significant performance degradation under noise. In this paper, we propose FCPE, a fast context-based pitch estimation model that employs a Lynx-Net architecture with depth-wise separable convolutions to effectively capture mel spectrogram features while maintaining low computational cost and robust noise tolerance. Experiments show that our method achieves 96.79\% Raw Pitch Accuracy (RPA) on the MIR-1K dataset, on par with the state-of-the-art methods. The Real-Time Factor (RTF) is 0.0062 on a single RTX 4090 GPU, which significantly outperforms existing algorithms in efficiency. Code is available at https://github.com/CNChTu/FCPE.

Choral singing, a widely practiced form of ensemble singing, lacks comprehensive datasets in the realm of Music Information Retrieval (MIR) research, due to challenges arising from the requirement to curate multitrack recordings. To address this, we devised a novel methodology, leveraging state-of-the-art synthesizers to create and curate quality renditions. The scores were sourced from Choral Public Domain Library(CPDL). This work is done in collaboration with a diverse team of musicians, software engineers and researchers. The resulting dataset, complete with its associated metadata, and methodology is released as part of this work, opening up new avenues for exploration and advancement in the field of singing voice research.

In speech generation tasks, human subjective ratings, usually referred to as the opinion score, are considered the "gold standard" for speech quality evaluation, with the mean opinion score (MOS) serving as the primary evaluation metric. Due to the high cost of human annotation, several MOS prediction systems have emerged in the speech domain, demonstrating good performance. These MOS prediction models are trained using annotations from previous speech-related challenges. However, compared to the speech domain, the singing domain faces data scarcity and stricter copyright protections, leading to a lack of high-quality MOS-annotated datasets for singing. To address this, we propose SingMOS, a high-quality and diverse MOS dataset for singing, covering a range of Chinese and Japanese datasets. These synthesized vocals are generated using state-of-the-art models in singing synthesis, conversion, or resynthesis tasks and are rated by professional annotators alongside real vocals. Data analysis demonstrates the diversity and reliability of our dataset. Additionally, we conduct further exploration on SingMOS, providing insights for singing MOS prediction and guidance for the continued expansion of SingMOS.

The virtual world is being established in which digital humans are created indistinguishable from real humans. Producing their audio-related capabilities is crucial since voice conveys extensive personal characteristics. We aim to create a controllable audio-form virtual singer; however, supervised modeling and controlling all different factors of the singing voice, such as timbre, tempo, pitch, and lyrics, is extremely difficult since accurately labeling all such information needs enormous labor work. In this paper, we propose a framework that could digitize a person's voice by simply "listening" to the clean voice recordings of any content in a fully unsupervised manner and predict singing voices even only using speaking recordings. A variational auto-encoder (VAE) based framework is developed, which leverages a set of pre-trained models to encode the audio as various hidden embeddings representing different factors of the singing voice, and further decodes the embeddings into raw audio. By manipulating the hidden embeddings for different factors, the resulting singing voices can be controlled, and new virtual singers can also be further generated by interpolating between timbres. Evaluations of different types of experiments demonstrate the proposed method's effectiveness. The proposed method is the critical technique for producing the AI choir, which empowered the human-AI symbiotic orchestra in Hong Kong in July 2022.

We introduce Jukebox, a model that generates music with singing in the raw audio domain. We tackle the long context of raw audio using a multi-scale VQ-VAE to compress it to discrete codes, and modeling those using autoregressive Transformers. We show that the combined model at scale can generate high-fidelity and diverse songs with coherence up to multiple minutes. We can condition on artist and genre to steer the musical and vocal style, and on unaligned lyrics to make the singing more controllable. We are releasing thousands of non cherry-picked samples at https://jukebox.openai.com, along with model weights and code at https://github.com/openai/jukebox

With the rise of multimodal large language models, GPT-4o stands out as a pioneering model, driving us to evaluate its capabilities. This report assesses GPT-4o across various tasks to analyze its audio processing and reasoning abilities. We find that GPT-4o exhibits strong knowledge in audio, speech, and music understanding, performing well in tasks like intent classification, spoken command classification, semantic and grammatical reasoning., multilingual speech recognition, and singing analysis. It also shows greater robustness against hallucinations than other large audio-language models (LALMs). However, it struggles with tasks such as audio duration prediction and instrument classification. Additionally, GPT-4o's safety mechanisms cause it to decline tasks like speaker identification, age classification, MOS prediction, and audio deepfake detection. Notably, the model exhibits a significantly different refusal rate when responding to speaker verification tasks on different datasets. This is likely due to variations in the accompanying instructions or the quality of the input audio, suggesting the sensitivity of its built-in safeguards. Finally, we acknowledge that model performance varies with evaluation protocols. This report only serves as a preliminary exploration of the current state of LALMs.

Various applications of voice synthesis have been developed independently despite the fact that they generate "voice" as output in common. In addition, the majority of voice synthesis models currently rely on annotated audio data, but it is crucial to scale them to self-supervised datasets in order to effectively capture the wide range of acoustic variations present in human voice, including speaker identity, emotion, and prosody. In this work, we propose Make-A-Voice, a unified framework for synthesizing and manipulating voice signals from discrete representations. Make-A-Voice leverages a "coarse-to-fine" approach to model the human voice, which involves three stages: 1) semantic stage: model high-level transformation between linguistic content and self-supervised semantic tokens, 2) acoustic stage: introduce varying control signals as acoustic conditions for semantic-to-acoustic modeling, and 3) generation stage: synthesize high-fidelity waveforms from acoustic tokens. Make-A-Voice offers notable benefits as a unified voice synthesis framework: 1) Data scalability: the major backbone (i.e., acoustic and generation stage) does not require any annotations, and thus the training data could be scaled up. 2) Controllability and conditioning flexibility: we investigate different conditioning mechanisms and effectively handle three voice synthesis applications, including text-to-speech (TTS), voice conversion (VC), and singing voice synthesis (SVS) by re-synthesizing the discrete voice representations with prompt guidance. Experimental results demonstrate that Make-A-Voice exhibits superior audio quality and style similarity compared with competitive baseline models. Audio samples are available at https://Make-A-Voice.github.io

Despite recent progress in generative adversarial network (GAN)-based vocoders, where the model generates raw waveform conditioned on acoustic features, it is challenging to synthesize high-fidelity audio for numerous speakers across various recording environments. In this work, we present BigVGAN, a universal vocoder that generalizes well for various out-of-distribution scenarios without fine-tuning. We introduce periodic activation function and anti-aliased representation into the GAN generator, which brings the desired inductive bias for audio synthesis and significantly improves audio quality. In addition, we train our GAN vocoder at the largest scale up to 112M parameters, which is unprecedented in the literature. We identify and address the failure modes in large-scale GAN training for audio, while maintaining high-fidelity output without over-regularization. Our BigVGAN, trained only on clean speech (LibriTTS), achieves the state-of-the-art performance for various zero-shot (out-of-distribution) conditions, including unseen speakers, languages, recording environments, singing voices, music, and instrumental audio. We release our code and model at: https://github.com/NVIDIA/BigVGAN

Singing voice detection (SVD), to recognize vocal parts in the song, is an essential task in music information retrieval (MIR). The task remains challenging since singing voice varies and intertwines with the accompaniment music, especially for some complicated polyphonic music such as choral music recordings. To address this problem, we investigate singing voice detection while discarding the interference from the accompaniment. The proposed SVD has two steps: i. The singing voice separation (SVS) technique is first utilized to filter out the singing voice's potential part coarsely. ii. Upon the continuity of vocal in the time domain, Long-term Recurrent Convolutional Networks (LRCN) is used to learn compositional features. Moreover, to eliminate the outliers, we choose to use a median filter for time-domain smoothing. Experimental results show that the proposed method outperforms the existing state-of-the-art works on two public datasets, the Jamendo Corpus and the RWC pop dataset.

Zero-shot voice conversion aims to transform a source speech utterance to match the timbre of a reference speech from an unseen speaker. Traditional approaches struggle with timbre leakage, insufficient timbre representation, and mismatches between training and inference tasks. We propose Seed-VC, a novel framework that addresses these issues by introducing an external timbre shifter during training to perturb the source speech timbre, mitigating leakage and aligning training with inference. Additionally, we employ a diffusion transformer that leverages the entire reference speech context, capturing fine-grained timbre features through in-context learning. Experiments demonstrate that Seed-VC outperforms strong baselines like OpenVoice and CosyVoice, achieving higher speaker similarity and lower word error rates in zero-shot voice conversion tasks. We further extend our approach to zero-shot singing voice conversion by incorporating fundamental frequency (F0) conditioning, resulting in comparative performance to current state-of-the-art methods. Our findings highlight the effectiveness of Seed-VC in overcoming core challenges, paving the way for more accurate and versatile voice conversion systems.

We introduce AnyEnhance, a unified generative model for voice enhancement that processes both speech and singing voices. Based on a masked generative model, AnyEnhance is capable of handling both speech and singing voices, supporting a wide range of enhancement tasks including denoising, dereverberation, declipping, super-resolution, and target speaker extraction, all simultaneously and without fine-tuning. AnyEnhance introduces a prompt-guidance mechanism for in-context learning, which allows the model to natively accept a reference speaker's timbre. In this way, it could boost enhancement performance when a reference audio is available and enable the target speaker extraction task without altering the underlying architecture. Moreover, we also introduce a self-critic mechanism into the generative process for masked generative models, yielding higher-quality outputs through iterative self-assessment and refinement. Extensive experiments on various enhancement tasks demonstrate AnyEnhance outperforms existing methods in terms of both objective metrics and subjective listening tests. Demo audios are publicly available at https://amphionspace.github.io/anyenhance/.

We propose a multi-singer emotional singing voice synthesizer, Muse-SVS, that expresses emotion at various intensity levels by controlling subtle changes in pitch, energy, and phoneme duration while accurately following the score. To control multiple style attributes while avoiding loss of fidelity and expressiveness due to interference between attributes, Muse-SVS represents all attributes and their relations together by a joint embedding in a unified embedding space. Muse-SVS can express emotional intensity levels not included in the training data through embedding interpolation and extrapolation. We also propose a statistical pitch predictor to express pitch variance according to emotional intensity, and a context-aware residual duration predictor to prevent the accumulation of variances in phoneme duration, which is crucial for synchronization with instrumental parts. In addition, we propose a novel ASPP-Transformer, which combines atrous spatial pyramid pooling (ASPP) and Transformer, to improve fidelity and expressiveness by referring to broad contexts. In experiments, Muse-SVS exhibited improved fidelity, expressiveness, and synchronization performance compared with baseline models. The visualization results show that Muse-SVS effectively express the variance in pitch, energy, and phoneme duration according to emotional intensity. To the best of our knowledge, Muse-SVS is the first neural SVS capable of controlling emotional intensity.

With recent advances in automatic speech recognition (ASR), large language models (LLMs), and text-to-speech (TTS) technologies, spoken dialogue systems (SDS) have become widely accessible. However, most existing SDS are limited to conventional spoken responses. We present SingingSDS, a cascaded SDS that responds through singing rather than speaking, fostering more affective, memorable, and pleasurable interactions in character-based roleplay and interactive entertainment scenarios. SingingSDS employs a modular ASR-LLM-SVS pipeline and supports a wide range of configurations across character personas, ASR and LLM backends, SVS models, melody sources, and voice profiles, tailored to different needs in terms of latency, quality, and musical style. SingingSDS is available as a plug-and-play web demo, featuring modular, open-source code that supports customization and extension. Demo: https://huggingface.co/spaces/espnet/SingingSDS. Code: https://github.com/SingingSDS/SingingSDS.

This paper presents our systems (denoted as T13) for the singing voice conversion challenge (SVCC) 2023. For both in-domain and cross-domain English singing voice conversion (SVC) tasks (Task 1 and Task 2), we adopt a recognition-synthesis approach with self-supervised learning-based representation. To achieve data-efficient SVC with a limited amount of target singer/speaker's data (150 to 160 utterances for SVCC 2023), we first train a diffusion-based any-to-any voice conversion model using publicly available large-scale 750 hours of speech and singing data. Then, we finetune the model for each target singer/speaker of Task 1 and Task 2. Large-scale listening tests conducted by SVCC 2023 show that our T13 system achieves competitive naturalness and speaker similarity for the harder cross-domain SVC (Task 2), which implies the generalization ability of our proposed method. Our objective evaluation results show that using large datasets is particularly beneficial for cross-domain SVC.

Singing voice conversion (SVC) aims to convert a singer's voice in a given music piece to another singer while keeping the original content. We propose an end-to-end feature disentanglement-based model, which we named SaMoye, to enable zero-shot many-to-many singing voice conversion. SaMoye disentangles the features of the singing voice into content features, timbre features, and pitch features respectively. The content features are enhanced using a GPT-based model to perform cross-prediction with the phoneme of the lyrics. SaMoye can generate the music with converted voice by replacing the timbre features with the target singer. We also establish an unparalleled large-scale dataset to guarantee zero-shot performance. The dataset consists of 1500k pure singing vocal clips containing at least 10,000 singers.

In this study, we present SingVisio, an interactive visual analysis system that aims to explain the diffusion model used in singing voice conversion. SingVisio provides a visual display of the generation process in diffusion models, showcasing the step-by-step denoising of the noisy spectrum and its transformation into a clean spectrum that captures the desired singer's timbre. The system also facilitates side-by-side comparisons of different conditions, such as source content, melody, and target timbre, highlighting the impact of these conditions on the diffusion generation process and resulting conversions. Through comprehensive evaluations, SingVisio demonstrates its effectiveness in terms of system design, functionality, explainability, and user-friendliness. It offers users of various backgrounds valuable learning experiences and insights into the diffusion model for singing voice conversion.

Singing-driven 3D head animation is a challenging yet promising task with applications in virtual avatars, entertainment, and education. Unlike speech, singing involves richer emotional nuance, dynamic prosody, and lyric-based semantics, requiring the synthesis of fine-grained, temporally coherent facial motion. Existing speech-driven approaches often produce oversimplified, emotionally flat, and semantically inconsistent results, which are insufficient for singing animation. To address this, we propose Think2Sing, a diffusion-based framework that leverages pretrained large language models to generate semantically coherent and temporally consistent 3D head animations, conditioned on both lyrics and acoustics. A key innovation is the introduction of motion subtitles, an auxiliary semantic representation derived through a novel Singing Chain-of-Thought reasoning process combined with acoustic-guided retrieval. These subtitles contain precise timestamps and region-specific motion descriptions, serving as interpretable motion priors. We frame the task as a motion intensity prediction problem, enabling finer control over facial regions and improving the modeling of expressive motion. To support this, we create a multimodal singing dataset with synchronized video, acoustic descriptors, and motion subtitles, enabling diverse and expressive motion learning. Extensive experiments show that Think2Sing outperforms state-of-the-art methods in realism, expressiveness, and emotional fidelity, while also offering flexible, user-controllable animation editing.

Audio generation systems now create very realistic soundscapes that can enhance media production, but also pose potential risks. Several studies have examined deepfakes in speech or singing voice. However, environmental sounds have different characteristics, which may make methods for detecting speech and singing deepfakes less effective for real-world sounds. In addition, existing datasets for environmental sound deepfake detection are limited in scale and audio types. To address this gap, we introduce EnvSDD, the first large-scale curated dataset designed for this task, consisting of 45.25 hours of real and 316.74 hours of fake audio. The test set includes diverse conditions to evaluate the generalizability, such as unseen generation models and unseen datasets. We also propose an audio deepfake detection system, based on a pre-trained audio foundation model. Results on EnvSDD show that our proposed system outperforms the state-of-the-art systems from speech and singing domains.

Automatic Pitch Correction (APC) enhances vocal recordings by aligning pitch deviations with the intended musical notes. However, existing APC systems either rely on reference pitches, which limits their practical applicability, or employ simple pitch estimation algorithms that often fail to preserve expressiveness and naturalness. We propose BERT-APC, a novel reference-free APC framework that corrects pitch errors while maintaining the natural expressiveness of vocal performances. In BERT-APC, a novel stationary pitch predictor first estimates the perceived pitch of each note from the detuned singing voice. A context-aware note pitch predictor estimates the intended pitch sequence by leveraging a music language model repurposed to incorporate musical context. Finally, a note-level correction algorithm fixes pitch errors while preserving intentional pitch deviations for emotional expression. In addition, we introduce a learnable data augmentation strategy that improves the robustness of the music language model by simulating realistic detuning patterns. Compared to two recent singing voice transcription models, BERT-APC demonstrated superior performance in note pitch prediction, outperforming the second-best model, ROSVOT, by 10.49%p on highly detuned samples in terms of the raw pitch accuracy. In the MOS test, BERT-APC achieved the highest score of 4.32 pm 0.15, which is significantly higher than those of the widely-used commercial APC tools, AutoTune (3.22 pm 0.18) and Melodyne (3.08 pm 0.18), while maintaining a comparable ability to preserve expressive nuances. To the best of our knowledge, this is the first APC model that leverages a music language model to achieve reference-free pitch correction with symbolic musical context. The corrected audio samples of BERT-APC are available online.

Singing voice beautifying is a novel task that has application value in people's daily life, aiming to correct the pitch of the singing voice and improve the expressiveness without changing the original timbre and content. Existing methods rely on paired data or only concentrate on the correction of pitch. However, professional songs and amateur songs from the same person are hard to obtain, and singing voice beautifying doesn't only contain pitch correction but other aspects like emotion and rhythm. Since we propose a fast and high-fidelity singing voice beautifying system called ConTuner, a diffusion model combined with the modified condition to generate the beautified Mel-spectrogram, where the modified condition is composed of optimized pitch and expressiveness. For pitch correction, we establish a mapping relationship from MIDI, spectrum envelope to pitch. To make amateur singing more expressive, we propose the expressiveness enhancer in the latent space to convert amateur vocal tone to professional. ConTuner achieves a satisfactory beautification effect on both Mandarin and English songs. Ablation study demonstrates that the expressiveness enhancer and generator-based accelerate method in ConTuner are effective.

Generative Adversarial Network (GAN) based vocoders are superior in inference speed and synthesis quality when reconstructing an audible waveform from an acoustic representation. This study focuses on improving the discriminator to promote GAN-based vocoders. Most existing time-frequency-representation-based discriminators are rooted in Short-Time Fourier Transform (STFT), whose time-frequency resolution in a spectrogram is fixed, making it incompatible with signals like singing voices that require flexible attention for different frequency bands. Motivated by that, our study utilizes the Constant-Q Transform (CQT), which owns dynamic resolution among frequencies, contributing to a better modeling ability in pitch accuracy and harmonic tracking. Specifically, we propose a Multi-Scale Sub-Band CQT (MS-SB-CQT) Discriminator, which operates on the CQT spectrogram at multiple scales and performs sub-band processing according to different octaves. Experiments conducted on both speech and singing voices confirm the effectiveness of our proposed method. Moreover, we also verified that the CQT-based and the STFT-based discriminators could be complementary under joint training. Specifically, enhanced by the proposed MS-SB-CQT and the existing MS-STFT Discriminators, the MOS of HiFi-GAN can be boosted from 3.27 to 3.87 for seen singers and from 3.40 to 3.78 for unseen singers.

In the domain of music and sound processing, pitch extraction plays a pivotal role. Our research presents a specialized convolutional neural network designed for pitch extraction, particularly from the human singing voice in acapella performances. Notably, our approach combines synthetic data with auto-labeled acapella sung audio, creating a robust training environment. Evaluation across datasets comprising synthetic sounds, opera recordings, and time-stretched vowels demonstrates its efficacy. This work paves the way for enhanced pitch extraction in both music and voice settings.

Representing speech and audio signals in discrete units has become a compelling alternative to traditional high-dimensional feature vectors. Numerous studies have highlighted the efficacy of discrete units in various applications such as speech compression and restoration, speech recognition, and speech generation. To foster exploration in this domain, we introduce the Interspeech 2024 Challenge, which focuses on new speech processing benchmarks using discrete units. It encompasses three pivotal tasks, namely multilingual automatic speech recognition, text-to-speech, and singing voice synthesis, and aims to assess the potential applicability of discrete units in these tasks. This paper outlines the challenge designs and baseline descriptions. We also collate baseline and selected submission systems, along with preliminary findings, offering valuable contributions to future research in this evolving field.

Singing voice synthesis is a generative task that involves multi-dimensional control of the singing model, including lyrics, pitch, and duration, and includes the timbre of the singer and singing skills such as vibrato. In this paper, we proposed SU-net for singing voice synthesis named SUSing. Synthesizing singing voice is treated as a translation task between lyrics and music score and spectrum. The lyrics and music score information is encoded into a two-dimensional feature representation through the convolution layer. The two-dimensional feature and its frequency spectrum are mapped to the target spectrum in an autoregressive manner through a SU-net network. Within the SU-net the stripe pooling method is used to replace the alternate global pooling method to learn the vertical frequency relationship in the spectrum and the changes of frequency in the time domain. The experimental results on the public dataset Kiritan show that the proposed method can synthesize more natural singing voices.

We present the findings of the latest iteration of the Singing Voice Conversion Challenge, a scientific event aiming to compare and understand different voice conversion systems in a controlled environment. Compared to previous iterations which solely focused on converting the singer identity, this year we also focused on converting the singing style of the singer. To create a controlled environment and thorough evaluations, we developed a new challenge database, introduced two tasks, open-sourced baselines, and conducted large-scale crowd-sourced listening tests and objective evaluations. The challenge was ran for two months and in total we evaluated 26 different systems. The results of the large-scale crowd-sourced listening test showed that top systems had comparable singer identity scores to ground truth samples. However, modeling the singing style and consequently achieving high naturalness still remains a challenge in this task, primarily due to the difficulty in modeling dynamic information in breathy, glissando, and vibrato singing styles.

This work presents FreeSVC, a promising multilingual singing voice conversion approach that leverages an enhanced VITS model with Speaker-invariant Clustering (SPIN) for better content representation and the State-of-the-Art (SOTA) speaker encoder ECAPA2. FreeSVC incorporates trainable language embeddings to handle multiple languages and employs an advanced speaker encoder to disentangle speaker characteristics from linguistic content. Designed for zero-shot learning, FreeSVC enables cross-lingual singing voice conversion without extensive language-specific training. We demonstrate that a multilingual content extractor is crucial for optimal cross-language conversion. Our source code and models are publicly available.

Separating a singing voice from its music accompaniment remains an important challenge in the field of music information retrieval. We present a unique neural network approach inspired by a technique that has revolutionized the field of vision: pixel-wise image classification, which we combine with cross entropy loss and pretraining of the CNN as an autoencoder on singing voice spectrograms. The pixel-wise classification technique directly estimates the sound source label for each time-frequency (T-F) bin in our spectrogram image, thus eliminating common pre- and postprocessing tasks. The proposed network is trained by using the Ideal Binary Mask (IBM) as the target output label. The IBM identifies the dominant sound source in each T-F bin of the magnitude spectrogram of a mixture signal, by considering each T-F bin as a pixel with a multi-label (for each sound source). Cross entropy is used as the training objective, so as to minimize the average probability error between the target and predicted label for each pixel. By treating the singing voice separation problem as a pixel-wise classification task, we additionally eliminate one of the commonly used, yet not easy to comprehend, postprocessing steps: the Wiener filter postprocessing. The proposed CNN outperforms the first runner up in the Music Information Retrieval Evaluation eXchange (MIREX) 2016 and the winner of MIREX 2014 with a gain of 2.2702 ~ 5.9563 dB global normalized source to distortion ratio (GNSDR) when applied to the iKala dataset. An experiment with the DSD100 dataset on the full-tracks song evaluation task also shows that our model is able to compete with cutting-edge singing voice separation systems which use multi-channel modeling, data augmentation, and model blending.

Vocal recordings on consumer devices commonly suffer from multiple concurrent degradations: noise, reverberation, band-limiting, and clipping. We present Smule Renaissance Small (SRS), a compact single-stage model that performs end-to-end vocal restoration directly in the complex STFT domain. By incorporating phase-aware losses, SRS enables large analysis windows for improved frequency resolution while achieving 10.5x real-time inference on iPhone 12 CPU at 48 kHz. On the DNS 5 Challenge blind set, despite no speech training, SRS outperforms a strong GAN baseline and closely matches a computationally expensive flow-matching system. To enable evaluation under realistic multi-degradation scenarios, we introduce the Extreme Degradation Bench (EDB): 87 singing and speech recordings captured under severe acoustic conditions. On EDB, SRS surpasses all open-source baselines on singing and matches commercial systems, while remaining competitive on speech despite no speech-specific training. We release both SRS and EDB under the MIT License.

Current leading Text-To-Audio (TTA) generation models suffer from degraded performance on zero-shot and few-shot settings. It is often challenging to generate high-quality audio for audio events that are unseen or uncommon in the training set. Inspired by the success of Retrieval-Augmented Generation (RAG) in Large Language Model (LLM)-based knowledge-intensive tasks, we extend the TTA process with additional conditioning contexts. We propose Audiobox TTA-RAG, a novel retrieval-augmented TTA approach based on Audiobox, a conditional flow-matching audio generation model. Unlike the vanilla Audiobox TTA solution which generates audio conditioned on text, we augmented the conditioning input with retrieved audio samples that provide additional acoustic information to generate the target audio. Our retrieval method does not require the external database to have labeled audio, offering more practical use cases. To evaluate our proposed method, we curated test sets in zero-shot and few-shot settings. Our empirical results show that the proposed model can effectively leverage the retrieved audio samples and significantly improve zero-shot and few-shot TTA performance, with large margins on multiple evaluation metrics, while maintaining the ability to generate semantically aligned audio for the in-domain setting. In addition, we investigate the effect of different retrieval methods and data sources.

Generating full-length, high-quality songs is challenging, as it requires maintaining long-term coherence both across text and music modalities and within the music modality itself. Existing non-autoregressive (NAR) frameworks, while capable of producing high-quality songs, often struggle with the alignment between lyrics and vocal. Concurrently, catering to diverse musical preferences necessitates reinforcement learning from human feedback (RLHF). However, existing methods often rely on merging multiple models during multi-preference optimization, which results in significant performance degradation. To address these challenges, we introduce DiffRhythm 2, an end-to-end framework designed for high-fidelity, controllable song generation. To tackle the lyric alignment problem, DiffRhythm 2 employs a semi-autoregressive architecture based on block flow matching. This design enables faithful alignment of lyrics to singing vocals without relying on external labels and constraints, all while preserving the high generation quality and efficiency of NAR models. To make this framework computationally tractable for long sequences, we implement a music variational autoencoder (VAE) that achieves a low frame rate of 5 Hz while still enabling high-fidelity audio reconstruction. In addition, to overcome the limitations of multi-preference optimization in RLHF, we propose cross-pair preference optimization. This method effectively mitigates the performance drop typically associated with model merging, allowing for more robust optimization across diverse human preferences. We further enhance musicality and structural coherence by introducing stochastic block representation alignment loss.

In this work, we study the task of multi-singer separation in a cappella music, where the number of active singers varies across mixtures. To address this, we use a power set-based data augmentation strategy that expands limited multi-singer datasets into exponentially more training samples. To separate singers, we introduce SepACap, an adaptation of SepReformer, a state-of-the-art speaker separation model architecture. We adapt the model with periodic activations and a composite loss function that remains effective when stems are silent, enabling robust detection and separation. Experiments on the JaCappella dataset demonstrate that our approach achieves state-of-the-art performance in both full-ensemble and subset singer separation scenarios, outperforming spectrogram-based baselines while generalizing to realistic mixtures with varying numbers of singers.

Joint audio-video generation aims to synthesize synchronized multisensory content, yet current unified models struggle with fine-grained acoustic control, particularly for identity-preserving speech. Existing approaches either suffer from temporal misalignment due to cascaded generation or lack the capability to perform zero-shot voice cloning within a joint synthesis framework. In this work, we present MM-Sonate, a multimodal flow-matching framework that unifies controllable audio-video joint generation with zero-shot voice cloning capabilities. Unlike prior works that rely on coarse semantic descriptions, MM-Sonate utilizes a unified instruction-phoneme input to enforce strict linguistic and temporal alignment. To enable zero-shot voice cloning, we introduce a timbre injection mechanism that effectively decouples speaker identity from linguistic content. Furthermore, addressing the limitations of standard classifier-free guidance in multimodal settings, we propose a noise-based negative conditioning strategy that utilizes natural noise priors to significantly enhance acoustic fidelity. Empirical evaluations demonstrate that MM-Sonate establishes new state-of-the-art performance in joint generation benchmarks, significantly outperforming baselines in lip synchronization and speech intelligibility, while achieving voice cloning fidelity comparable to specialized Text-to-Speech systems.

The goal of universal audio representation learning is to obtain foundational models that can be used for a variety of downstream tasks involving speech, music or environmental sounds. To approach this problem, methods inspired by self-supervised models from NLP, like BERT, are often used and adapted to audio. These models rely on the discrete nature of text, hence adopting this type of approach for audio processing requires either a change in the learning objective or mapping the audio signal to a set of discrete classes. In this work, we explore the use of EnCodec, a neural audio codec, to generate discrete targets for learning an universal audio model based on a masked autoencoder (MAE). We evaluate this approach, which we call EncodecMAE, on a wide range of audio tasks spanning speech, music and environmental sounds, achieving performances comparable or better than leading audio representation models.

An automatic pitch correction system typically includes several stages, such as pitch extraction, deviation estimation, pitch shift processing, and cross-fade smoothing. However, designing these components with strategies often requires domain expertise and they are likely to fail on corner cases. In this paper, we present KaraTuner, an end-to-end neural architecture that predicts pitch curve and resynthesizes the singing voice directly from the tuned pitch and vocal spectrum extracted from the original recordings. Several vital technical points have been introduced in KaraTuner to ensure pitch accuracy, pitch naturalness, timbre consistency, and sound quality. A feed-forward Transformer is employed in the pitch predictor to capture longterm dependencies in the vocal spectrum and musical note. We also develop a pitch-controllable vocoder based on a novel source-filter block and the Fre-GAN architecture. KaraTuner obtains a higher preference than the rule-based pitch correction approach through A/B tests, and perceptual experiments show that the proposed vocoder achieves significant advantages in timbre consistency and sound quality compared with the parametric WORLD vocoder, phase vocoder and CLPC vocoder.

This paper explores the potential universality of neural vocoders. We train a WaveRNN-based vocoder on 74 speakers coming from 17 languages. This vocoder is shown to be capable of generating speech of consistently good quality (98% relative mean MUSHRA when compared to natural speech) regardless of whether the input spectrogram comes from a speaker or style seen during training or from an out-of-domain scenario when the recording conditions are studio-quality. When the recordings show significant changes in quality, or when moving towards non-speech vocalizations or singing, the vocoder still significantly outperforms speaker-dependent vocoders, but operates at a lower average relative MUSHRA of 75%. These results are shown to be consistent across languages, regardless of them being seen during training (e.g. English or Japanese) or unseen (e.g. Wolof, Swahili, Ahmaric).

Singing Accompaniment Generation (SAG) is the process of generating instrumental music for a given clean vocal input. However, existing SAG techniques use source-separated vocals as input and overfit to separation artifacts. This creates a critical train-test mismatch, leading to failure on clean, real-world vocal inputs. We introduce AnyAccomp, a framework that resolves this by decoupling accompaniment generation from source-dependent artifacts. AnyAccomp first employs a quantized melodic bottleneck, using a chromagram and a VQ-VAE to extract a discrete and timbre-invariant representation of the core melody. A subsequent flow-matching model then generates the accompaniment conditioned on these robust codes. Experiments show AnyAccomp achieves competitive performance on separated-vocal benchmarks while significantly outperforming baselines on generalization test sets of clean studio vocals and, notably, solo instrumental tracks. This demonstrates a qualitative leap in generalization, enabling robust accompaniment for instruments - a task where existing models completely fail - and paving the way for more versatile music co-creation tools. Demo audio and code: https://anyaccomp.github.io

Text-to-audio generation (TTA) produces audio from a text description, learning from pairs of audio samples and hand-annotated text. However, commercializing audio generation is challenging as user-input prompts are often under-specified when compared to text descriptions used to train TTA models. In this work, we treat TTA models as a ``blackbox'' and address the user prompt challenge with two key insights: (1) User prompts are generally under-specified, leading to a large alignment gap between user prompts and training prompts. (2) There is a distribution of audio descriptions for which TTA models are better at generating higher quality audio, which we refer to as ``audionese''. To this end, we rewrite prompts with instruction-tuned models and propose utilizing text-audio alignment as feedback signals via margin ranking learning for audio improvements. On both objective and subjective human evaluations, we observed marked improvements in both text-audio alignment and music audio quality.

Discrete audio tokens have recently gained attention for their potential to bridge the gap between audio and language processing. Ideal audio tokens must preserve content, paralinguistic elements, speaker identity, and many other audio details. Current audio tokenization methods fall into two categories: Semantic tokens, acquired through quantization of Self-Supervised Learning (SSL) models, and Neural compression-based tokens (codecs). Although previous studies have benchmarked codec models to identify optimal configurations, the ideal setup for quantizing pretrained SSL models remains unclear. This paper explores the optimal configuration of semantic tokens across discriminative and generative tasks. We propose a scalable solution to train a universal vocoder across multiple SSL layers. Furthermore, an attention mechanism is employed to identify task-specific influential layers, enhancing the adaptability and performance of semantic tokens in diverse audio applications.

Music is an integral part of human culture, embodying human intelligence and creativity, of which songs compose an essential part. While various aspects of song generation have been explored by previous works, such as singing voice, vocal composition and instrumental arrangement, etc., generating songs with both vocals and accompaniment given lyrics remains a significant challenge, hindering the application of music generation models in the real world. In this light, we propose SongCreator, a song-generation system designed to tackle this challenge. The model features two novel designs: a meticulously designed dual-sequence language model (DSLM) to capture the information of vocals and accompaniment for song generation, and an additional attention mask strategy for DSLM, which allows our model to understand, generate and edit songs, making it suitable for various song-related generation tasks. Extensive experiments demonstrate the effectiveness of SongCreator by achieving state-of-the-art or competitive performances on all eight tasks. Notably, it surpasses previous works by a large margin in lyrics-to-song and lyrics-to-vocals. Additionally, it is able to independently control the acoustic conditions of the vocals and accompaniment in the generated song through different prompts, exhibiting its potential applicability. Our samples are available at https://songcreator.github.io/.

Human voice encodes both identity and paralinguistic cues, yet encoders in large audio-language models (LALMs) rarely balance both aspects. In this work, we present a study toward building a general-purpose voice encoder that captures nuanced voice cues. Through a comprehensive evaluation, we find that multi-task training yields the most balanced representations, whereas contrastive language-audio pretraining (CLAP) primarily improves retrieval without enhancing paralinguistic understanding. Our final encoder, Auden-Voice, also demonstrates strong performance when integrated with LLMs. The code and training recipes will be released with the audio understanding toolkit Auden.

Recognizing human non-speech vocalizations is an important task and has broad applications such as automatic sound transcription and health condition monitoring. However, existing datasets have a relatively small number of vocal sound samples or noisy labels. As a consequence, state-of-the-art audio event classification models may not perform well in detecting human vocal sounds. To support research on building robust and accurate vocal sound recognition, we have created a VocalSound dataset consisting of over 21,000 crowdsourced recordings of laughter, sighs, coughs, throat clearing, sneezes, and sniffs from 3,365 unique subjects. Experiments show that the vocal sound recognition performance of a model can be significantly improved by 41.9% by adding VocalSound dataset to an existing dataset as training material. In addition, different from previous datasets, the VocalSound dataset contains meta information such as speaker age, gender, native language, country, and health condition.

Automatic lyrics transcription (ALT) remains a challenging task in the field of music information retrieval, despite great advances in automatic speech recognition (ASR) brought about by transformer-based architectures in recent years. One of the major challenges in ALT is the high amplitude of interfering audio signals relative to conventional ASR due to musical accompaniment. Recent advances in music source separation have enabled automatic extraction of high-quality separated vocals, which could potentially improve ALT performance. However, the effect of source separation has not been systematically investigated in order to establish best practices for its use. This work examines the impact of source separation on ALT using Whisper, a state-of-the-art open source ASR model. We evaluate Whisper's performance on original audio, separated vocals, and vocal stems across short-form and long-form transcription tasks. For short-form, we suggest a concatenation method that results in a consistent reduction in Word Error Rate (WER). For long-form, we propose an algorithm using source separation as a vocal activity detector to derive segment boundaries, which results in a consistent reduction in WER relative to Whisper's native long-form algorithm. Our approach achieves state-of-the-art results for an open source system on the Jam-ALT long-form ALT benchmark, without any training or fine-tuning. We also publish MUSDB-ALT, the first dataset of long-form lyric transcripts following the Jam-ALT guidelines for which vocal stems are publicly available.

We propose Fast Language-Audio Pre-training (FLAP), a self-supervised approach that efficiently and effectively learns aligned audio and language representations through masking, contrastive learning and reconstruction. For efficiency, FLAP randomly drops audio spectrogram tokens, focusing solely on the remaining ones for self-supervision. Through inter-modal contrastive learning, FLAP learns to align paired audio and text representations in a shared latent space. Notably, FLAP leverages multiple augmented views via masking for inter-modal contrast and learns to reconstruct the masked portion of audio tokens. Moreover, FLAP leverages large language models (LLMs) to augment the text inputs, contributing to improved performance. These approaches lead to more robust and informative audio-text representations, enabling FLAP to achieve state-of-the-art (SoTA) performance on audio-text retrieval tasks on AudioCaps (achieving 53.0% R@1) and Clotho (achieving 25.5% R@1).

We propose WHISPER-GPT: A generative large language model (LLM) for speech and music that allows us to work with continuous audio representations and discrete tokens simultaneously as part of a single architecture. There has been a huge surge in generative audio, speech, and music models that utilize discrete audio tokens derived from neural compression algorithms, e.g. ENCODEC. However, one of the major drawbacks of this approach is handling the context length. It blows up for high-fidelity generative architecture if one has to account for all the audio contents at various frequencies for the next token prediction. By combining continuous audio representation like the spectrogram and discrete acoustic tokens, we retain the best of both worlds: Have all the information needed from the audio at a specific time instance in a single token, yet allow LLM to predict the future token to allow for sampling and other benefits discrete space provides. We show how our architecture improves the perplexity and negative log-likelihood scores for the next token prediction compared to a token-based LLM for speech and music.

Singing voice conversion (SVC) is one promising technique which can enrich the way of human-computer interaction by endowing a computer the ability to produce high-fidelity and expressive singing voice. In this paper, we propose DiffSVC, an SVC system based on denoising diffusion probabilistic model. DiffSVC uses phonetic posteriorgrams (PPGs) as content features. A denoising module is trained in DiffSVC, which takes destroyed mel spectrogram produced by the diffusion/forward process and its corresponding step information as input to predict the added Gaussian noise. We use PPGs, fundamental frequency features and loudness features as auxiliary input to assist the denoising process. Experiments show that DiffSVC can achieve superior conversion performance in terms of naturalness and voice similarity to current state-of-the-art SVC approaches.

We propose a flexible framework that deals with both singer conversion and singers vocal technique conversion. The proposed model is trained on non-parallel corpora, accommodates many-to-many conversion, and leverages recent advances of variational autoencoders. It employs separate encoders to learn disentangled latent representations of singer identity and vocal technique separately, with a joint decoder for reconstruction. Conversion is carried out by simple vector arithmetic in the learned latent spaces. Both a quantitative analysis as well as a visualization of the converted spectrograms show that our model is able to disentangle singer identity and vocal technique and successfully perform conversion of these attributes. To the best of our knowledge, this is the first work to jointly tackle conversion of singer identity and vocal technique based on a deep learning approach.

Recent advances in large language models (LLMs) and audio language models have significantly improved music generation, particularly in lyrics-to-song generation. However, existing approaches still struggle with the complex composition of songs and the scarcity of high-quality data, leading to limitations in sound quality, musicality, instruction following, and vocal-instrument harmony. To address these challenges, we introduce LeVo, an LM-based framework consisting of LeLM and a music codec. LeLM is capable of parallelly modeling two types of tokens: mixed tokens, which represent the combined audio of vocals and accompaniment to achieve vocal-instrument harmony, and dual-track tokens, which separately encode vocals and accompaniment for high-quality song generation. It employs two decoder-only transformers and a modular extension training strategy to prevent interference between different token types. To further enhance musicality and instruction following, we introduce a multi-preference alignment method based on Direct Preference Optimization (DPO). This method handles diverse human preferences through a semi-automatic data construction process and DPO post-training. Experimental results demonstrate that LeVo consistently outperforms existing methods on both objective and subjective metrics. Ablation studies further justify the effectiveness of our designs. Audio examples are available at https://levo-demo.github.io/.

Song generation focuses on producing controllable high-quality songs based on various prompts. However, existing methods struggle to generate vocals and accompaniments with prompt-based control and proper alignment. Additionally, they fall short in supporting various tasks. To address these challenges, we introduce VersBand, a multi-task song generation framework for synthesizing high-quality, aligned songs with prompt-based control. VersBand comprises these primary models: 1) VocalBand, a decoupled model, leverages the flow-matching method for generating singing styles, pitches, and mel-spectrograms, allowing fast, high-quality vocal generation with style control. 2) AccompBand, a flow-based transformer model, incorporates the Band-MOE, selecting suitable experts for enhanced quality, alignment, and control. This model allows for generating controllable, high-quality accompaniments aligned with vocals. 3) Two generation models, LyricBand for lyrics and MelodyBand for melodies, contribute to the comprehensive multi-task song generation system, allowing for extensive control based on multiple prompts. Experimental results demonstrate that VersBand performs better over baseline models across multiple song generation tasks using objective and subjective metrics. Audio samples are available at https://VersBand.github.io.

Neural codec language models have achieved state-of-the-art performance in text-to-speech (TTS) synthesis, leveraging scalable architectures like autoregressive transformers and large-scale speech datasets. By framing voice cloning as a prompt continuation task, these models excel at cloning voices from short audio samples. However, this approach is limited in its ability to handle numerous or lengthy speech excerpts, since the concatenation of source and target speech must fall within the maximum context length which is determined during training. In this work, we introduce Lina-Speech, a model that replaces traditional self-attention mechanisms with emerging recurrent architectures like Gated Linear Attention (GLA). Building on the success of initial-state tuning on RWKV, we extend this technique to voice cloning, enabling the use of multiple speech samples and full utilization of the context window in synthesis. This approach is fast, easy to deploy, and achieves performance comparable to fine-tuned baselines when the dataset size ranges from 3 to 15 minutes. Notably, Lina-Speech matches or outperforms state-of-the-art baseline models, including some with a parameter count up to four times higher or trained in an end-to-end style. We release our code and checkpoints. Audio samples are available at https://theodorblackbird.github.io/blog/demo_lina/.

Rap, a prominent genre of vocal performance, remains underexplored in vocal generation. General vocal synthesis depends on precise note and duration inputs, requiring users to have related musical knowledge, which limits flexibility. In contrast, rap typically features simpler melodies, with a core focus on a strong rhythmic sense that harmonizes with accompanying beats. In this paper, we propose Freestyler, the first system that generates rapping vocals directly from lyrics and accompaniment inputs. Freestyler utilizes language model-based token generation, followed by a conditional flow matching model to produce spectrograms and a neural vocoder to restore audio. It allows a 3-second prompt to enable zero-shot timbre control. Due to the scarcity of publicly available rap datasets, we also present RapBank, a rap song dataset collected from the internet, alongside a meticulously designed processing pipeline. Experimental results show that Freestyler produces high-quality rapping voice generation with enhanced naturalness and strong alignment with accompanying beats, both stylistically and rhythmically.

Producing synthetic voice, similar to human-like sound, is an emerging novelty of modern interactive media systems. Text-To-Speech (TTS) systems try to generate synthetic and authentic voices via text input. Besides, well known and familiar dubbing, announcing and narrating voices, as valuable possessions of any media organization, can be kept forever by utilizing TTS and Voice Conversion (VC) algorithms . The emergence of deep learning approaches has made such TTS systems more accurate and accessible. To understand TTS systems better, this paper investigates the key components of such systems including text analysis, acoustic modelling and vocoding. The paper then provides details of important state-of-the-art TTS systems based on deep learning. Finally, a comparison is made between recently released systems in term of backbone architecture, type of input and conversion, vocoder used and subjective assessment (MOS). Accordingly, Tacotron 2, Transformer TTS, WaveNet and FastSpeech 1 are among the most successful TTS systems ever released. In the discussion section, some suggestions are made to develop a TTS system with regard to the intended application.

Recent breakthroughs in zero-shot voice synthesis have enabled imitating a speaker's voice using just a few seconds of recording while maintaining a high level of realism. Alongside its potential benefits, this powerful technology introduces notable risks, including voice fraud and speaker impersonation. Unlike the conventional approach of solely relying on passive methods for detecting synthetic data, watermarking presents a proactive and robust defence mechanism against these looming risks. This paper introduces an innovative audio watermarking framework that encodes up to 32 bits of watermark within a mere 1-second audio snippet. The watermark is imperceptible to human senses and exhibits strong resilience against various attacks. It can serve as an effective identifier for synthesized voices and holds potential for broader applications in audio copyright protection. Moreover, this framework boasts high flexibility, allowing for the combination of multiple watermark segments to achieve heightened robustness and expanded capacity. Utilizing 10 to 20-second audio as the host, our approach demonstrates an average Bit Error Rate (BER) of 0.48\% across ten common attacks, a remarkable reduction of over 2800\% in BER compared to the state-of-the-art watermarking tool. See https://aka.ms/wavmark for demos of our work.

We introduce Matcha-TTS, a new encoder-decoder architecture for speedy TTS acoustic modelling, trained using optimal-transport conditional flow matching (OT-CFM). This yields an ODE-based decoder capable of high output quality in fewer synthesis steps than models trained using score matching. Careful design choices additionally ensure each synthesis step is fast to run. The method is probabilistic, non-autoregressive, and learns to speak from scratch without external alignments. Compared to strong pre-trained baseline models, the Matcha-TTS system has the smallest memory footprint, rivals the speed of the fastest models on long utterances, and attains the highest mean opinion score in a listening test. Please see https://shivammehta25.github.io/Matcha-TTS/ for audio examples, code, and pre-trained models.

Diffusion-based audio and music generation models commonly generate music by constructing an image representation of audio (e.g., a mel-spectrogram) and then converting it to audio using a phase reconstruction model or vocoder. Typical vocoders, however, produce monophonic audio at lower resolutions (e.g., 16-24 kHz), which limits their effectiveness. We propose MusicHiFi -- an efficient high-fidelity stereophonic vocoder. Our method employs a cascade of three generative adversarial networks (GANs) that convert low-resolution mel-spectrograms to audio, upsamples to high-resolution audio via bandwidth expansion, and upmixes to stereophonic audio. Compared to previous work, we propose 1) a unified GAN-based generator and discriminator architecture and training procedure for each stage of our cascade, 2) a new fast, near downsampling-compatible bandwidth extension module, and 3) a new fast downmix-compatible mono-to-stereo upmixer that ensures the preservation of monophonic content in the output. We evaluate our approach using both objective and subjective listening tests and find our approach yields comparable or better audio quality, better spatialization control, and significantly faster inference speed compared to past work. Sound examples are at https://MusicHiFi.github.io/web/.

The cloning of a speaker's voice using an untranscribed reference sample is one of the great advances of modern neural text-to-speech (TTS) methods. Approaches for mimicking the prosody of a transcribed reference audio have also been proposed recently. In this work, we bring these two tasks together for the first time through utterance level normalization in conjunction with an utterance level speaker embedding. We further introduce a lightweight aligner for extracting fine-grained prosodic features, that can be finetuned on individual samples within seconds. We show that it is possible to clone the voice of a speaker as well as the prosody of a spoken reference independently without any degradation in quality and high similarity to both original voice and prosody, as our objective evaluation and human study show. All of our code and trained models are available, alongside static and interactive demos.

Vocal pitch is an important high-level feature in music audio processing. However, extracting vocal pitch in polyphonic music is more challenging due to the presence of accompaniment. To eliminate the influence of the accompaniment, most previous methods adopt music source separation models to obtain clean vocals from polyphonic music before predicting vocal pitches. As a result, the performance of vocal pitch estimation is affected by the music source separation models. To address this issue and directly extract vocal pitches from polyphonic music, we propose a robust model named RMVPE. This model can extract effective hidden features and accurately predict vocal pitches from polyphonic music. The experimental results demonstrate the superiority of RMVPE in terms of raw pitch accuracy (RPA) and raw chroma accuracy (RCA). Additionally, experiments conducted with different types of noise show that RMVPE is robust across all signal-to-noise ratio (SNR) levels. The code of RMVPE is available at https://github.com/Dream-High/RMVPE.

Recent work has shown that it is possible to resynthesize high-quality speech based, not on text, but on low bitrate discrete units that have been learned in a self-supervised fashion and can therefore capture expressive aspects of speech that are hard to transcribe (prosody, voice styles, non-verbal vocalization). The adoption of these methods is still limited by the fact that most speech synthesis datasets are read, severely limiting spontaneity and expressivity. Here, we introduce Expresso, a high-quality expressive speech dataset for textless speech synthesis that includes both read speech and improvised dialogues rendered in 26 spontaneous expressive styles. We illustrate the challenges and potentials of this dataset with an expressive resynthesis benchmark where the task is to encode the input in low-bitrate units and resynthesize it in a target voice while preserving content and style. We evaluate resynthesis quality with automatic metrics for different self-supervised discrete encoders, and explore tradeoffs between quality, bitrate and invariance to speaker and style. All the dataset, evaluation metrics and baseline models are open source

This paper presents VoiceLDM, a model designed to produce audio that accurately follows two distinct natural language text prompts: the description prompt and the content prompt. The former provides information about the overall environmental context of the audio, while the latter conveys the linguistic content. To achieve this, we adopt a text-to-audio (TTA) model based on latent diffusion models and extend its functionality to incorporate an additional content prompt as a conditional input. By utilizing pretrained contrastive language-audio pretraining (CLAP) and Whisper, VoiceLDM is trained on large amounts of real-world audio without manual annotations or transcriptions. Additionally, we employ dual classifier-free guidance to further enhance the controllability of VoiceLDM. Experimental results demonstrate that VoiceLDM is capable of generating plausible audio that aligns well with both input conditions, even surpassing the speech intelligibility of the ground truth audio on the AudioCaps test set. Furthermore, we explore the text-to-speech (TTS) and zero-shot text-to-audio capabilities of VoiceLDM and show that it achieves competitive results. Demos and code are available at https://voiceldm.github.io.

Despite recent progress in text-to-audio (TTA) generation, we show that the state-of-the-art models, such as AudioLDM, trained on datasets with an imbalanced class distribution, such as AudioCaps, are biased in their generation performance. Specifically, they excel in generating common audio classes while underperforming in the rare ones, thus degrading the overall generation performance. We refer to this problem as long-tailed text-to-audio generation. To address this issue, we propose a simple retrieval-augmented approach for TTA models. Specifically, given an input text prompt, we first leverage a Contrastive Language Audio Pretraining (CLAP) model to retrieve relevant text-audio pairs. The features of the retrieved audio-text data are then used as additional conditions to guide the learning of TTA models. We enhance AudioLDM with our proposed approach and denote the resulting augmented system as Re-AudioLDM. On the AudioCaps dataset, Re-AudioLDM achieves a state-of-the-art Frechet Audio Distance (FAD) of 1.37, outperforming the existing approaches by a large margin. Furthermore, we show that Re-AudioLDM can generate realistic audio for complex scenes, rare audio classes, and even unseen audio types, indicating its potential in TTA tasks.

This paper studies a simple extension of image-based Masked Autoencoders (MAE) to self-supervised representation learning from audio spectrograms. Following the Transformer encoder-decoder design in MAE, our Audio-MAE first encodes audio spectrogram patches with a high masking ratio, feeding only the non-masked tokens through encoder layers. The decoder then re-orders and decodes the encoded context padded with mask tokens, in order to reconstruct the input spectrogram. We find it beneficial to incorporate local window attention in the decoder, as audio spectrograms are highly correlated in local time and frequency bands. We then fine-tune the encoder with a lower masking ratio on target datasets. Empirically, Audio-MAE sets new state-of-the-art performance on six audio and speech classification tasks, outperforming other recent models that use external supervised pre-training. The code and models will be at https://github.com/facebookresearch/AudioMAE.

Time-aligned lyrics can enrich the music listening experience by enabling karaoke, text-based song retrieval and intra-song navigation, and other applications. Compared to text-to-speech alignment, lyrics alignment remains highly challenging, despite many attempts to combine numerous sub-modules including vocal separation and detection in an effort to break down the problem. Furthermore, training required fine-grained annotations to be available in some form. Here, we present a novel system based on a modified Wave-U-Net architecture, which predicts character probabilities directly from raw audio using learnt multi-scale representations of the various signal components. There are no sub-modules whose interdependencies need to be optimized. Our training procedure is designed to work with weak, line-level annotations available in the real world. With a mean alignment error of 0.35s on a standard dataset our system outperforms the state-of-the-art by an order of magnitude.

Hindustani music is a performance-driven oral tradition that exhibits the rendition of rich melodic patterns. In this paper, we focus on generative modeling of singers' vocal melodies extracted from audio recordings, as the voice is musically prominent within the tradition. Prior generative work in Hindustani music models melodies as coarse discrete symbols which fails to capture the rich expressive melodic intricacies of singing. Thus, we propose to use a finely quantized pitch contour, as an intermediate representation for hierarchical audio modeling. We propose GaMaDHaNi, a modular two-level hierarchy, consisting of a generative model on pitch contours, and a pitch contour to audio synthesis model. We compare our approach to non-hierarchical audio models and hierarchical models that use a self-supervised intermediate representation, through a listening test and qualitative analysis. We also evaluate audio model's ability to faithfully represent the pitch contour input using Pearson correlation coefficient. By using pitch contours as an intermediate representation, we show that our model may be better equipped to listen and respond to musicians in a human-AI collaborative setting by highlighting two potential interaction use cases (1) primed generation, and (2) coarse pitch conditioning.

Self-supervised learning (SSL) has recently emerged as a promising paradigm for training generalisable models on large-scale data in the fields of vision, text, and speech. Although SSL has been proven effective in speech and audio, its application to music audio has yet to be thoroughly explored. This is primarily due to the distinctive challenges associated with modelling musical knowledge, particularly its tonal and pitched characteristics of music. To address this research gap, we propose an acoustic Music undERstanding model with large-scale self-supervised Training (MERT), which incorporates teacher models to provide pseudo labels in the masked language modelling (MLM) style acoustic pre-training. In our exploration, we identified a superior combination of teacher models, which outperforms conventional speech and audio approaches in terms of performance. This combination includes an acoustic teacher based on Residual Vector Quantization - Variational AutoEncoder (RVQ-VAE) and a musical teacher based on the Constant-Q Transform (CQT). These teachers effectively guide our student model, a BERT-style transformer encoder, to better model music audio. In addition, we introduce an in-batch noise mixture augmentation to enhance the representation robustness. Furthermore, we explore a wide range of settings to overcome the instability in acoustic language model pre-training, which allows our designed paradigm to scale from 95M to 330M parameters. Experimental results indicate that our model can generalise and perform well on 14 music understanding tasks and attains state-of-the-art (SOTA) overall scores. The code and models are online: https://github.com/yizhilll/MERT.

Audio Language Models (ALM) have emerged as the dominant paradigm for speech and music generation by representing audio as sequences of discrete tokens. Yet, unlike text tokens, which are invertible, audio tokens are extracted from lossy codecs with a limited bitrate. As a consequence, increasing audio quality requires generating more tokens, which imposes a trade-off between fidelity and computational cost. We address this issue by studying Continuous Audio Language Models (CALM). These models instantiate a large Transformer backbone that produces a contextual embedding at every timestep. This sequential information then conditions an MLP that generates the next continuous frame of an audio VAE through consistency modeling. By avoiding lossy compression, CALM achieves higher quality at lower computational cost than their discrete counterpart. Experiments on speech and music demonstrate improved efficiency and fidelity over state-of-the-art discrete audio language models, facilitating lightweight, high-quality audio generation. Samples are available at https://continuous-audio-language-models.github.io