app.py

import gradio as gr
from transformers import AutoModel
import soundfile as sf
import numpy as np

# Load IndicF5 model
model = AutoModel.from_pretrained("ai4bharat/IndicF5", trust_remote_code=True)

def generate_tts(text, ref_audio, ref_text):
    if ref_audio is None:
        return "⚠️ कृपया एक आवाज फाइल अपलोड करा", None

    # ref_audio = path string (because gr.Audio gives filepath)
    ref_audio_path = ref_audio

    # Run model
    audio = model(text, ref_audio_path=ref_audio_path, ref_text=ref_text)
    audio = np.array(audio, dtype=np.float32)

    # Save output
    out_path = "output.wav"
    sf.write(out_path, audio, samplerate=24000)

    return "✅ आवाज तयार झाला!", out_path

with gr.Blocks() as demo:
    gr.Markdown("## 🎙️ माझं Marathi Voice Cloning (IndicF5)")

    text = gr.Textbox(label="तुम्हाला काय बोलायचं आहे?")
    ref_audio = gr.Audio(type="filepath", label="तुमचा आवाज (WAV, 6–30 सेकंद)")
    ref_text = gr.Textbox(label="त्या ऑडिओमध्ये तुम्ही काय बोललात?")
    out_msg = gr.Textbox(label="Status")
    out_audio = gr.Audio(label="Generated Audio", type="filepath")

    btn = gr.Button("Generate Speech")
    btn.click(generate_tts, inputs=[text, ref_audio, ref_text], outputs=[out_msg, out_audio])

demo.launch()