Commit History
-
feat(separation): Integrate BS-RoFormer & Mel-RoFormer models
This commit introduces support for two additional audio separation models, BS-RoFormer and Mel-RoFormer, providing users with more specialized options for vocal and instrumental separation.d050f96
-
feat(output): Add parameter metadata to final audio files
Saves the complete set of `AppParameters` used for each render into the metadata of the output FLAC file. This makes each render self-documenting and allows users to easily track the settings used for any given result. The parameters are stored as a JSON string in the file's `comment` tag.1db4ef0
-
feat: Add dual-stem MIDI transcription and `basic-pitch` profiles
**1. Dual-Stem MIDI Transcription:** - When using vocal separation, a new option allows for transcribing **both** the vocal and accompaniment stems independently. - The two resulting MIDI files are then automatically merged into a single, more complete MIDI file for rendering. This improves transcription quality on complex tracks. **2. `basic-pitch` Profile Presets:** - A "Transcription Profile Preset" dropdown has been added to the UI for the general-purpose transcription method. - Includes a library of presets optimized for different instruments (Vocals, Piano, Drums) and genres (Rock, Jazz, Classical). - Selecting a profile automatically configures all `basic-pitch` parameters for better results on specific audio types.2923df9
-
Add experimental vocal separation and selective transcription features
- Added **Separate Vocals** option (experimental): Splits input audio into vocals and music stems before processing. - Added **Transcription Target** setting: Allows choosing which stem (vocals or music) to transcribe to MIDI. - Added option to **Re-merge Other Part with Rendered Audio**: After rendering, merges the non-transcribed stem (e.g., original vocals) back with the new music.22dd15a
-
feat: Implement stereo audio to MIDI transcription
This commit introduces a new stereo processing workflow for audio-to-MIDI transcription, allowing the preservation of spatial information from stereo recordings. The previous implementation was limited to mono processing. Scale MIDI velocities by 0.8 in Stereo Transcription to avoid loudness/clipping after merge Applied `scale_instrument_velocity(scale=0.8)` during Stereo Transcription to prevent excessive loudness caused by summing left and right channel MIDI tracks. This helps maintain a more natural dynamic range, avoiding clipping and ensuring more consistent perceived volume after rendering to WAV/FLAC.b114cd4
-
Integrated and Enhanced: Audio-to-MIDI and Advanced MIDI Renderer
- **Integration of asigalov61's Projects:** Merges the `ByteDance-Solo-Piano-Audio-to-MIDI-Transcription` and `Advanced-MIDI-Renderer` projects into a single, cohesive application. This forms the foundational workflow for MIDI processing and rendering. - **Addition of General-Purpose Transcription:** Implements Spotify's `basic-pitch` library as a new transcription option. This allows for the effective transcription of non-piano music, including songs with vocals and multiple instruments. - **Dynamic UI for Method Selection:** A new UI control allows users to choose between the "General Purpose" (basic-pitch) and "Piano-Specific" (ByteDance) transcription models, with the UI dynamically showing relevant settings for the selected method. - **Refactor: Extract shared MIDI module to reduce duplication:** The `midi_to_colab_audio.py` and `TMIDIX.py` scripts both contained a large, duplicated block of code for MIDI processing. - **Improve FluidSynth rendering speed in midi_to_colab_audi…adcbc9f
-
Update requirements.txt
Verified signature988c0de
-
Update requirements.txt
Verified signatured559da8
-
Update requirements.txt
Verified signature9d71a06 -
Update requirements.txt
Verified signature169c4f3
-
Update requirements.txt
5efa190
-
Create requirements.txt
9bff65c