main.py

import time

from fastapi import FastAPI, UploadFile
from fastapi.middleware.cors import CORSMiddleware


import torch
from transformers import pipeline

app = FastAPI(docs_url="/api/docs")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
    allow_credentials=True,
)

device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

BATCH_SIZE = 8


pipe = pipeline("automatic-speech-recognition",
                "openai/whisper-large-v3",
                torch_dtype=torch_dtype,
                device=device)


@app.get("/device")
def getDevice():
    start_time = time.time()
    print("Time took to process the request and return response is {} sec".format(
        time.time() - start_time))
    return device


@app.post("/transcribe")
def transcribe(soundFile: UploadFile, task="transcribe"):
    start_time = time.time()

    if soundFile is None:
        raise "No audio file submitted! Please upload or record an audio file before submitting your request."

    inputFile = soundFile.file.read()

    text = pipe(inputFile, batch_size=BATCH_SIZE, generate_kwargs={
                "task": task}, return_timestamps=True)["text"]

    print("Time took to process the request and return response is {} sec".format(
        time.time() - start_time))
    return text