app.py

import torch
import gradio as gr
import pytube as pt
from transformers import pipeline
from fastspeech2 import FastSpeech2

MODEL_NAME = "openai/whisper-large-v2"

device = 0 if torch.cuda.is_available() else "cpu"

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

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

voice_conversion_model = FastSpeech2.from_pretrained("path/to/pretrained/voice_conversion_model")

def convert_voice(text):
    converted_voice = voice_conversion_model(text)
    return converted_voice

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"]
    converted_voice = convert_voice(text)

    return state + "\n" + converted_voice, state + "\n" + converted_voice, converted_voice

mf_transcribe = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.Audio(source="microphone", type="filepath", optional=True),
        gr.State(value="")
    ],
    outputs=[
        gr.Textbox(lines=15),
        gr.State(),
        gr.Audio(type="auto")  # Add this line to include the converted voice as an output
    ],
    layout="horizontal",
    theme="huggingface",
    title="Whisper Large V2: Transcribe Audio and Voice Conversion",
    live=True,
    description=(
        "Transcribe long-form microphone or audio inputs and convert the voice with the click of a button! Demo uses the"
        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
        " of arbitrary length and FastSpeech2 for voice conversion."
    ),
    allow_flagging="never",
)

mf_transcribe.launch(enable_queue=True)