app.py

import gradio as gr
import spaces
import torch
import io
import os.path
import urllib.request
import whisper

os.makedirs("/home/user/.cache/whisper")
urllib.request.urlretrieve(
    "https://openaipublic.azureedge.net/main/whisper/models/d7440d1dc186f76616474e0ff0b3b6b879abc9d1a4926b7adfa41db2d497ab4f/medium.en.pt",
    "/home/user/.cache/whisper/medium.en.pt"
)

writer = whisper.utils.get_writer("srt", "/dev/null")

@spaces.GPU
def generate(file, progress=gr.Progress(track_tqdm=True)):
    # get file to type bytes somehow
    model = whisper.load_model("medium.en", device="cuda")
    audio = whisper.load_audio(file)
    result = model.transcribe(audio, verbose=False)
    out = io.StringIO()
    writer.write_result(result, out)
    return out.getvalue()

gr.Interface(fn=generate, inputs=gr.File(type="filepath"), outputs=gr.Text()).launch()