app.py

from fastapi import FastAPI, HTTPException
from fastapi.responses import FileResponse
from pydantic import BaseModel
import torch
from transformers import AutoTokenizer, AutoModelForTextToWaveform
from scipy.io.wavfile import write as wav_write
import tempfile
import os

app = FastAPI()

# Initialize the TTS model globally
tok = AutoTokenizer.from_pretrained("facebook/mms-tts-som")
model = AutoModelForTextToWaveform.from_pretrained("facebook/mms-tts-som")


class TextRequest(BaseModel):
    text: str


@app.post("/tts")
async def text_to_speech(request: TextRequest):
    try:
        if not request.text.strip():
            raise HTTPException(status_code=400, detail="Text cannot be empty")

        # Generate audio
        inputs = tok(request.text, return_tensors="pt")

        with torch.no_grad():
            audio = model(**inputs).waveform

        sr = model.config.sampling_rate  # usually 16000 Hz
        audio_numpy = audio.squeeze().cpu().numpy()

        # Speed up audio by resampling (increase sampling rate)
        speed_factor = 1.09  # Adjust this value: 1.0 = normal, 1.5 = 50% faster, 2.0 = 2x faster
        new_sr = int(sr * speed_factor)

        # Create temporary file
        with tempfile.NamedTemporaryFile(delete=False, suffix='.wav') as tmp_file:
            wav_write(tmp_file.name, new_sr, audio_numpy)

            # Return the file
            return FileResponse(
                tmp_file.name,
                media_type='audio/wav',
                filename='output.wav'
            )

    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))


@app.get("/health")
async def health_check():
    return {"status": "healthy", "model": "facebook/mms-tts-som"}

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)