app.py

import gradio as gr
import whisper
from pytube import YouTube
from typing import List
from transformers import pipeline

def transcribe(
        url: str,
        model_size: str
) -> str:
    # Get audio from the video.
    yt_client = YouTube(url=url)
    audio_file = yt_client.streams.filter(only_audio=True)[0].download(filename="file.mp4")

    # Load the model
    model = whisper.load_model(model_size)

    # Load the audio into the model
    audio = whisper.load_audio(audio_file)

    # Get results
    result = model.transcribe(audio)
    return format_result(result), summarize(result["text"])


def summarize(text: str) -> str:
    summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
    out = summarizer(text, max_length=150, min_length=30, do_sample=False)[0]["summary_text"]
    return out


def format_result(result: whisper.DecodingResult) -> str:
    out = []
    for item in result["segments"]:
        out.append(f"from {item['start']:6.2f} to {item['end']:6.2f} {item['text']}")
    return "\n".join(out)


def get_model_sizes() -> List[str]:
    """
    :rtype: list
    :return: List of possible sizes of the Whisper model.
    """
    return list(
        whisper._MODELS.keys()
    )


title = "YouTube transcribe + summarization"
desc = "Transcribe YouTube videos using OpenAI Whisper."

with gr.Blocks() as demo:
    gr.HTML(title)
    with gr.Row():
        with gr.Column():
            gr.Markdown(
                f"""
                {desc}
                """
            )
    with gr.Row():
        model_size = gr.Dropdown(
            label="Model size",
            choices=get_model_sizes(),
            value="tiny"
        )
        url = gr.Textbox(label="YouTube URL")
    with gr.Row():
        text = gr.Textbox(
            label="Transcription",
            lines=10
        )
    with gr.Row():
        summarization = gr.Textbox(
            label="Summarization",
            lines=5
        )
    with gr.Row().style(equal_height=True):
        submit_button = gr.Button("Submit")

    submit_button.click(
        transcribe,
        inputs=[
            url,
            model_size
        ],
        outputs=[
            text,
            summarization
        ]
    )

demo.launch()