app.py

import gradio as gr
import torch
import soundfile as sf
from transformers import pipeline, SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
import pdfplumber
import tqdm
import soundfile as sf


# Initialize summarization pipeline globally (load once)
summarizer = pipeline("summarization", model="t5-large")

def extract_abstract(path):
    with pdfplumber.open(path) as pdf:
        text = ""
        for page in tqdm.tqdm(pdf.pages):
            txt = page.extract_text(x_tolerance=2)
            if txt:
                text += txt
        abstract_start = text.find("Abstract")
        if text.find("Introduction") != -1:
            abstract_end = text.find("Introduction")
        else:
            abstract_end = text.find("Contents")
        abstract = text[abstract_start+len("Abstract"):abstract_end].strip()
    return abstract

def summarize_text(text):
    # Add prompt to encourage single sentence summary
    prompt = "Summarize the following text in one sentence: " + text
    result = summarizer(prompt, max_length=68, min_length=40, do_sample=False, num_beams=1, early_stopping=True)
    summary = result[0]['summary_text'].replace(" .", ",").replace(" . ", ", ").strip()
    if not summary.endswith("."):
        summary += "."
    return summary

def pdf_to_speech(file):
    # Gradio file input provides file.name as local path
    abstract = extract_abstract(file.name)
    summary = summarize_text(abstract)

    processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
    model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts")
    vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")

    inputs = processor(text=summary, return_tensors="pt")
    speaker_embeddings = torch.zeros(1, 512)  # Neutral speaker embedding

    speech = model.generate_speech(inputs["input_ids"], speaker_embeddings, vocoder=vocoder)
    speech_np = speech.squeeze().cpu().numpy()

    wav_output_path = "output.wav"
    sf.write(wav_output_path, speech_np, 16000)

    # Return path of wav file for Gradio audio output and download
    return wav_output_path

# Create Gradio interface with file input and audio output
demo = gr.Interface(
    fn=pdf_to_speech,
    inputs=gr.File(file_types=[".pdf"]),
    outputs=gr.Audio(type="filepath", label="Generated Audio"),
    title="PDF to Speech summary",
    description="Upload a PDF and get a spoken summary audio output."
)

demo.launch(debug=True, prevent_thread_lock=True, share=True)