app.py

import gradio as gr
import torch
import PyPDF2
from transformers import pipeline 
import numpy
import scipy
from gtts import gTTS
from io import BytesIO
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

def extract_text(pdf_file):
    pdfReader = PyPDF2.PdfReader(pdf_file)
    pageObj = pdfReader.pages[0]
    return pageObj.extract_text()

def summarize_text(text):
    sentences = text.split(". ")
    
    # Find abstract section
    for i, sentence in enumerate(sentences):
        if "Abstract" in sentence:
            start = i + 1
            end = start + 3
            break
            
    # Extract abstract 
    abstract = ". ".join(sentences[start:end+1])

    # Load BART model & tokenizer
    tokenizer = AutoTokenizer.from_pretrained("facebook/bart-large-cnn")
    model = AutoModelForSeq2SeqLM.from_pretrained("facebook/bart-large-cnn", low_cpu_mem_usage=True)

    # Tokenize abstract 
    inputs = tokenizer(abstract, return_tensors="pt", truncation=True)

    # Generate summary
    summary_ids = model.generate(inputs['input_ids'], num_beams=5, max_length=45, min_length=30, early_stopping=True)
    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)

    return summary

def text_to_audio(text):
    tts = gTTS(text, lang='en') 
    buffer = BytesIO()
    tts.write_to_fp(buffer)
    buffer.seek(0)  
    return buffer.read()

def audio_pdf(pdf_file):
    text = extract_text(pdf_file)
    summary = summarize_text(text) 
    audio = text_to_audio(summary)
    return summary, audio

inputs = gr.File() 
summary_text = gr.Text()
audio_summary = gr.Audio()


iface = gr.Interface(
    fn=audio_pdf,
    inputs=inputs,
    outputs=[summary_text,audio_summary],
    title="PDF Audio Summarizer 📻",
    description="App that converts an abstract into audio",
    examples=["Attention_is_all_you_need.pdf", 
              "ImageNet_Classification.pdf"
             ]
)

iface.launch()