app.py

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

transcribe_model_ckpt = "openai/whisper-small"
lang = "en"

transcribe_pipe = pipeline(
    task="automatic-speech-recognition",
    model=transcribe_model_ckpt,
    chunk_length_s=30,
)
transcribe_pipe.model.config.forced_decoder_ids = transcribe_pipe.tokenizer.get_decoder_prompt_ids(language=lang, task="transcribe")

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

    return html_embed_str, text

qa_model_ckpt = "deepset/tinyroberta-squad2"
qa_pipe = pipeline('question-answering', model=qa_model_ckpt, tokenizer=qa_model_ckpt)

def get_answer(query,context):
    QA_input = {
    'question': query,
    'context': context
    }
    res = qa_pipe(QA_input)["answer"]
    return res



with gr.Blocks() as demo:
    gr.Markdown("<h1><center>Youtube-QA</center></h1>")
    gr.Markdown("<h3>Ask questions from your youtube video of choice</h3>")
    gr.Markdown("""mermaid
                    graph LR
                    A[Youtube-audio] -->B(openai-whisper)
                    B -->C(Trascription)
                    C -->|Query| D(QA-model)
                    D -->E[Answer]
                    """)
    with gr.Column():
        with gr.Row():
            in_yt = gr.inputs.Textbox(lines=1, placeholder="Enter Youtube URL", label="YouTube URL")
        with gr.Row():
            transcribe_btn = gr.Button("Trascribe")
    with gr.Column():
        with gr.Row():
            out_yt_html = gradio.outputs.HTML()
        with gr.Row():
            out_yt_text = gradio.Textbox()
    with gr.Column():
        with gr.Row():
            in_query = gr.Textbox(lines=1, placeholder="What's your Question", label="Query")
        with gr.Row():
            ans_btn = gr.Button("Answer")
        with gr.Row():
            out_query = gradio.outputs.Textbox()
            
    
    transcribe_btn.click(fn=yt_transcribe, inputs=in_yt, outputs=[out_yt_html,out_yt_text])
    ans_btn.click(fn=get_answer, inputs=[in_query,out_yt_text], outputs=out_query)

demo.launch()