app.py

import gradio as gr
import spaces
import torch
from transformers.pipelines.audio_utils import ffmpeg_read
import io
import whisper

model = whisper.load_model("large", device="cuda")
writer = whisper.utils.get_writer("srt", "/dev/null")

@spaces.GPU
def generate(file):
    # get file to type bytes somehow
    audio = ffmpeg_read(file)
    result = model.transcribe(audio)
    out = io.StringIO()
    writer.write_result(result, out)
    return out.getvalue()

gr.Interface(fn=greet, inputs=gr.File(type="binary"), outputs=gr.Text()).launch()