demo-space/app.py

#!/usr/bin/env python
# -*- encoding: utf-8 -*-
'''
@Contact :   liuyuqi.gov@msn.cn
@Time    :   2023/05/12 18:19:33
@License :   Copyright © 2017-2022 liuyuqi. All Rights Reserved.
@Desc    :   whisper voice to text
'''

import torch
from transformers import pipeline
import gradio as gr
# import pytube as pt

MODE_NAME="openai/whisper-large-v2"
device= 0 if torch.cuda.is_available() else "cpu"

pipe = pipeline("automatic-speech-recognition", model=MODE_NAME, device=device,
chunk_length_s=30)


all_special_ids = pipe.tokenizer.all_special_ids
transcribe_token_id = all_special_ids[-5]
translate_token_id = all_special_ids[-6]


def transcribe(microphone, state, task="transcribe"):
    file = microphone
    pipe.model.config.forced_decoder_ids = [[2, transcribe_token_id if task=="transcribe" else translate_token_id]]
    text = pipe(file)["text"]
    return state + "\n" + text, state + "\n" + text

iface = gr.Interface(fn=transcribe, 
    inputs=[
        gr.Audio(source="microphone", type="filepath", optional=True),
        gr.State(value="")
    ], outputs=[
        gr.Textbox(lines=15),
        gr.State()],
    title="Speech to Text",
    layout="horizontal",
    theme="huggingface",
    live=True,
    description="Transcribe speech from your microphone or from a youtube video",
    allow_flagging="never",
    )
iface.launch(enable_queue=True)