app/main.py

from fastapi import FastAPI
import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline

app = FastAPI()

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

model_id = "openai/whisper-large-v3"

model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True,
    cache_dir="./"
)
model.to(device)

processor = AutoProcessor.from_pretrained(model_id, cache_dir="./")

pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    max_new_tokens=128,
    chunk_length_s=30,
    batch_size=16,
    return_timestamps=True,
    torch_dtype=torch_dtype,
    device=device,
)

def speech_to_text(path_to_file: str) -> str:
    """
    Given the path to the .wav file, it returns the transcription.
    """
    result = pipe(path_to_file)
    
    return result["text"]


@app.get("/")
def root():
    return {"Hello": "World"}

@app.post("/upload/")
async def create_upload_file(file: UploadFile = File(...)):
    
    # Process the file here
    content = await file.read()
    
    with open("tempfile.wav", "wb") as f:
        f.write(contents)
    response_text = speech_to_text("tempfile.wav")
    
    return PlainTextResponse(content=response_text, status_code=200)