app.py

from faster_whisper import WhisperModel
from fastapi import FastAPI
import requests
import os
import time

model_size = "large-v2"

# Run on GPU with FP16
model = WhisperModel(model_size, device="cuda", compute_type="float16")


app = FastAPI()


def handle(mp3):
    start_time = time.time()
    segments, info = model.transcribe(mp3, beam_size=1)
    end_time = time.time()
    target=[]
    for segment in segments:
        target.append({"start":segment.start,"end":segment.end,"text":segment.text})

    return {"start_at":start_time,"result":target,"end_at":end_time,"cost":(end_time-start_time)}


@app.get("/api/v3/trans")
async def download_mp3(url: str):
    r = requests.get(url, allow_redirects=True)
    mp3_file_path = "/tmp/" + url.split("/")[-1]
    with open(mp3_file_path, 'wb') as f:
        f.write(r.content)
    result = handle(mp3_file_path)
    return {"status": "success", "message": "mp3 file processed successfully","data":result}
```