app.py

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


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

# Load ASR model (Whisper)
asr_model = whisper.load_model("medium")

def generate_tts_and_transcribe(text, ref_audio, ref_text):
    # Save uploaded ref_audio to a path
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as tmp:
        tmp.write(ref_audio.read())
        ref_audio_path = tmp.name

    # Generate speech using IndicF5
    audio = tts_model(text, ref_audio_path=ref_audio_path, ref_text=ref_text)

    # Normalize
    if audio.dtype == np.int16:
        audio = audio.astype(np.float32) / 32768.0

    # Save TTS output
    tts_path = tempfile.NamedTemporaryFile(suffix=".wav", delete=False).name
    sf.write(tts_path, np.array(audio, dtype=np.float32), samplerate=24000)

    # Transcribe using Whisper
    asr_result = asr_model.transcribe(tts_path, language="ta")
    transcript = asr_result["text"]

    return tts_path, transcript

# Gradio Interface
demo = gr.Interface(
    fn=generate_tts_and_transcribe,
    inputs=[
        gr.Textbox(label="Text to Synthesize (Tamil)"),
        gr.Audio(label="Reference Audio (.wav)", type="file"),
        gr.Textbox(label="Reference Text (Tamil)")
    ],
    outputs=[
        gr.Audio(label="Generated Audio", type="filepath"),
        gr.Textbox(label="ASR Transcription (Whisper)")
    ],
    title="IndicF5 Tamil TTS + Whisper ASR",
    description="Give a reference audio and text, synthesize Tamil speech using IndicF5, and transcribe it with Whisper."
)

if __name__ == "__main__":
    demo.launch()