app.py

from fastapi import FastAPI, UploadFile, File
from fastapi.responses import JSONResponse
import whisperx
import torch
import tempfile
import shutil
import os

app = FastAPI()

# Load model globally to avoid reloading for every request
device = "cuda" if torch.cuda.is_available() else "cpu"
model = whisperx.load_model("medium", device)

@app.post("/transcribe")
async def transcribe_audio(file: UploadFile = File(...)):
    try:
        # Save uploaded audio to temp file
        with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
            shutil.copyfileobj(file.file, tmp)
            temp_audio_path = tmp.name

        # Load and process audio
        audio = whisperx.load_audio(temp_audio_path)
        result = model.transcribe(audio, batch_size=16, return_word_timestamps=True)

        # Clean up temp file
        os.remove(temp_audio_path)

        return JSONResponse(content=result)
    except Exception as e:
        return JSONResponse(status_code=500, content={"error": str(e)})