app.py

import os
import subprocess
import sys
import whisper
import torch
from transformers import pipeline
from transformers.utils import logging
from langdetect import detect
import gradio as gr
from gtts import gTTS
from moviepy.editor import VideoFileClip
import yt_dlp

# Set logging verbosity
logging.set_verbosity_error()

# Load the pre-trained Whisper model
whispermodel = whisper.load_model("medium")

# Load the summarizer pipeline
summarizer = pipeline(task="summarization", model="facebook/bart-large-cnn", torch_dtype=torch.bfloat16)

# Load the translator pipeline
translator = pipeline(task="translation", model="facebook/nllb-200-distilled-600M")

# Define language mappings
languages = {
    "English": "eng_Latn",
    "Arabic": "arb_Arab",
}

# Load QA pipeline
qa_pipeline = pipeline(task="question-answering", model="deepset/roberta-base-squad2")


# Function to extract audio from video
def extract_audio_from_video(video_file, output_audio="extracted_audio.mp3"):
    try:
        with VideoFileClip(video_file) as video_clip:
            video_clip.audio.write_audiofile(output_audio)
        return output_audio
    except Exception as e:
        return f"Error extracting audio: {e}"

# Define global variables
transcription = None
languageG = None

def content_input_update(content_type):
    visibility_map = {
        "Audio Upload": (True, False),
        "Video Upload": (False, True),
    }
    visible_audio, visible_video = visibility_map.get(content_type, (False, False))
    return (
        gr.update(visible=visible_audio),
        gr.update(visible=visible_video)
    )

def transcribe_content(content_type, audio_path, video):
    if content_type == "Audio Upload" and audio_path:
        return whispermodel.transcribe(audio_path)["text"]
    elif content_type == "Video Upload" and video:
        audio_file = extract_audio_from_video(video.name)
        return whispermodel.transcribe(audio_file)["text"]
    return None

def generate_summary(summarize):
    summary_text = None

    if summarize:
        summary = summarizer(transcription, min_length=10, max_length=150)
        summary_text = summary[0]['summary_text']

    return summary_text

def translator_text(summary, language):
    if language == 'English':
        return summary

    translated_summary = None
    translated_data = []

    if summary is not None:
        translated_summary = translator(summary, src_lang=languages["English"], tgt_lang=languages[language])[0]['translation_text']
    else:
        translated_summary = "No summary requested."

    return translated_summary

def create_audio_summary(summary, language):
    if summary and summary != 'No summary requested.':
        tts = gTTS(text=summary, lang='ar' if language == 'Arabic' else 'en')
        audio_path = "output_audio.mp3"
        tts.save(audio_path)
        return audio_path
    return None

def main(content_type, audio_path, video, language, summarize):
    global transcription, languageG
    languageG = language

    transcription = transcribe_content(content_type, audio_path, video)
    if not transcription:
        return "No transcription available.", "No Q&A requested.", None

    input_language = detect(transcription)
    input_language = 'Arabic' if input_language == 'ar' else 'English'
    if input_language != 'English':
        transcription = translator(transcription, src_lang=languages[input_language], tgt_lang=languages['English'])[0]['translation_text']

    summary_text = generate_summary(summarize)
    summary = translator_text(summary_text, language)
    audio_path = create_audio_summary(summary, language)
    
    return summary, audio_path

# Gradio interface
with gr.Blocks() as demo:
    gr.Markdown(
        """
        # Student Helper App
        This app allows students to upload audio, video, or YouTube links for automatic transcription.
        It can translate content, summarize it, and generate Q&A questions to help with studying.
        The app is ideal for students who want to review lectures, study materials, or any educational content more efficiently.
        """
    )

    content_type = gr.Radio(
        choices=["Audio Upload", "Video Upload"],
        label="Select Content Type",
        value="Audio Upload"
    )

    file_input = gr.Audio(label="Upload an Audio File", visible=True, type="filepath")
    video_input = gr.File(label="Upload a Video", visible=False, type="filepath")

    language = gr.Radio(choices=["Arabic", "English"], label="Preferred Language", value="English")
    summarize = gr.Checkbox(label="Summarize the content?")

    examples = [
        ["Audio Upload", "audio-example.mp3",  None, "English", True, True, 5],
        ["Video Upload", None, "video-example.mp4", "Arabic", True, False, 3],
    ]
    gr.Examples(
        examples=examples,
        inputs=[content_type, file_input, video_input, language, summarize],
        label="Try These Examples"
    )

    with gr.Tab("Summary"):
        summary_output = gr.Textbox(label="Summary", interactive=False)
        audio_output = gr.Audio(label="Audio Summary")

    with gr.Tab("Interactive Q&A"):
        user_question = gr.Textbox(label="Ask a Question", placeholder="Enter your question here...")
        qa_button = gr.Button("Get Answer")
        qa_response = gr.Markdown(label="Answer")

        qa_button.click(lambda question: interactive_qa(question), inputs=[user_question], outputs=qa_response)

    content_type.change(content_input_update, inputs=[content_type], outputs=[file_input, video_input])
    submit_btn = gr.Button("Submit")
    submit_btn.click(main, inputs=[content_type, file_input, video_input, language, summarize],
                      outputs=[summary_output, audio_output])

demo.launch(share=True)