app.py

import torch

import gradio as gr
import pytube as pt
from transformers import pipeline
from huggingface_hub import model_info

MODEL_NAME = "pierreguillou/whisper-medium-portuguese" #this always needs to stay in line 8 :D sorry for the hackiness
lang = "pt"

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

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

pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language=lang, task="transcribe")

def transcribe(microphone, file_upload):
    warn_output = ""
    if (microphone is not None) and (file_upload is not None):
        warn_output = (
            "WARNING: You've uploaded an audio file and used the microphone. "
            "The recorded file from the microphone will be used and the uploaded audio will be discarded.\n"
        )

    elif (microphone is None) and (file_upload is None):
        return "ERROR: You have to either use the microphone or upload an audio file"

    file = microphone if microphone is not None else file_upload

    text = pipe(file)["text"]

    return warn_output + text


def _return_yt_html_embed(yt_url):
    video_id = yt_url.split("?v=")[-1]
    HTML_str = (
        f'<center> <iframe width="500" height="320" src="https://www.youtube.com/embed/{video_id}"> </iframe>'
        " </center>"
    )
    return HTML_str


def yt_transcribe(yt_url):
    yt = pt.YouTube(yt_url)
    html_embed_str = _return_yt_html_embed(yt_url)
    stream = yt.streams.filter(only_audio=True)[0]
    stream.download(filename="audio.mp3")

    text = pipe("audio.mp3")["text"]

    return html_embed_str, text


demo = gr.Blocks()

mf_transcribe = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.inputs.Audio(source="microphone", type="filepath", optional=True),
        gr.inputs.Audio(source="upload", type="filepath", optional=True),
    ],
    outputs="text",
    layout="horizontal",
    theme="huggingface",
    title="Whisper Demo: Transcribe Portuguese Audio",
    description=(
        "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the the fine-tuned"
        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
        " of arbitrary length (all information about this Web APP in this blog post: [Speech-to-Text & IA | Transcreva qualquer áudio para o português com o Whisper (OpenAI)... sem nenhum custo!](https://medium.com/@pierre_guillou/speech-to-text-ia-transcreva-qualquer-%C3%A1udio-para-o-portugu%C3%AAs-com-o-whisper-openai-sem-ad0c17384681))."
    ),
    allow_flagging="never",
)

yt_transcribe = gr.Interface(
    fn=yt_transcribe,
    inputs=[gr.inputs.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL")],
    outputs=["html", "text"],
    layout="horizontal",
    theme="huggingface",
    title="Whisper Demo: Transcribe Portuguese YouTube",
    description=(
        "Transcribe long-form YouTube videos with the click of a button! Demo uses the the fine-tuned checkpoint:"
        f" [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files of"
        " arbitrary length (all information about this Web APP in this blog post: [Speech-to-Text & IA | Transcreva qualquer áudio para o português com o Whisper (OpenAI)... sem nenhum custo!](https://medium.com/@pierre_guillou/speech-to-text-ia-transcreva-qualquer-%C3%A1udio-para-o-portugu%C3%AAs-com-o-whisper-openai-sem-ad0c17384681))."
    ),
    allow_flagging="never",
)

with demo:
    gr.TabbedInterface([mf_transcribe, yt_transcribe], ["Transcribe Audio", "Transcribe YouTube"])

demo.launch(enable_queue=True)