app.py

import streamlit as st
from transformers import pipeline, AutoModelForSeq2SeqLM, AutoTokenizer
from TTS.api import TTS
from tempfile import NamedTemporaryFile
import os

# Initialize models
@st.cache_resource
def load_models():
    summarizer_model = AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-base")
    summarizer_tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-base")
    question_generator = pipeline("text2text-generation", model="valhalla/t5-small-qg-hl")
    qa_model = pipeline("question-answering", model="deepset/roberta-base-squad2")
    tts = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False, gpu=False)
    return summarizer_model, summarizer_tokenizer, question_generator, qa_model, tts

summarizer_model, summarizer_tokenizer, question_generator, qa_model, tts = load_models()

def summarize_document(text):
    input_ids = summarizer_tokenizer.encode(text, return_tensors="pt", max_length=512, truncation=True)
    summary_ids = summarizer_model.generate(input_ids, max_length=150, min_length=30, length_penalty=2.0, num_beams=4, early_stopping=True)
    return summarizer_tokenizer.decode(summary_ids[0], skip_special_tokens=True)

def generate_questions(summary):
    questions = question_generator(summary)
    return [q["generated_text"] for q in questions]

def generate_audio(text, voice_gender):
    voice = "ljspeech"  # Use default LJSpeech voice for both genders
    audio_file = NamedTemporaryFile(delete=False, suffix=".wav")
    tts.tts_to_file(text=text, file_path=audio_file.name)
    return audio_file.name

def app():
    st.title("Interactive Document Summarizer")
    uploaded_file = st.file_uploader("Upload a document", type=["txt", "pdf", "docx"])
    if uploaded_file:
        raw_text = uploaded_file.read().decode("utf-8")
        st.write("Processing document...")

        # Summarize
        summary = summarize_document(raw_text)
        st.write("Summary Generated:")
        st.write(summary)

        # Generate dialogue
        questions = generate_questions(summary)
        dialogue = []
        for idx, question in enumerate(questions):
            dialogue.append({"persona": "male" if idx % 2 == 0 else "female", "text": question})

        # Interactive simulation
        st.write("Simulating Conversation:")
        for item in dialogue:
            st.write(f"{item['persona'].capitalize()} says: {item['text']}")
            audio_path = generate_audio(item["text"], item["persona"])
            st.audio(audio_path, format="audio/wav")

if __name__ == "__main__":
    app()