Paper Voice

Convert academic papers to high-quality audio narration with precise mathematical explanations using a simplified LLM-powered approach.

Features

• 🧮 Natural Math Narration: Professor-style explanations of mathematical expressions
• 📄 Multi-Format Support: PDFs, LaTeX, Markdown, and plain text with math notation
• 🎯 Simple LLM Enhancement: Single comprehensive prompt for natural audio conversion
• 🗣️ Multiple TTS Options: OpenAI TTS (with chunking) or offline pyttsx3
• 💻 Web Interface: Easy-to-use Streamlit web app
• ⚡ Intelligent Chunking: Handles large documents with smart OpenAI API limits

Installation

From PyPI (Recommended)

pip install paper_voice

From Source

git clone https://github.com/gojiplus/paper_voice.git
cd paper_voice
pip install -e .

Usage

Web Interface (Recommended)

streamlit run paper_voice/streamlit/app.py

Upload a PDF, LaTeX file, or enter text directly. Provide an OpenAI API key for LLM-enhanced natural language conversion of mathematical expressions.

Python API

Simple Enhancement (New in v0.3.0)

from paper_voice.simple_llm_enhancer import enhance_document_simple

# Convert any academic content with math to natural language
content = "The equation $E = mc^2$ represents energy-mass equivalence."
enhanced = enhance_document_simple(content, api_key="your-openai-key")
print(enhanced)
# Output: "The equation energy equals mass times the speed of light squared represents energy-mass equivalence."

Complete Workflow

from paper_voice import pdf_utils
from paper_voice.simple_llm_enhancer import enhance_document_simple
from paper_voice import tts

# 1. Extract text from PDF
pages = pdf_utils.extract_raw_text("paper.pdf")
content = '\n\n'.join(pages)

# 2. Enhance with LLM (converts math to natural language)
enhanced_script = enhance_document_simple(content, api_key="your-openai-key")

# 3. Generate audio
tts.synthesize_speech_chunked(
    enhanced_script, 
    "output.mp3", 
    use_openai=True, 
    api_key="your-openai-key"
)

LaTeX Processing

from paper_voice.content_processor import process_content_unified

latex_content = r"""
\documentclass{article}
\begin{document}
The algorithm minimizes $J(\theta) = \frac{1}{2m}\sum_{i=1}^{m}(h_\theta(x^{(i)}) - y^{(i)})^2$.
\end{document}
"""

processed = process_content_unified(
    content=latex_content,
    input_type='latex',
    api_key='your-openai-key',
    use_llm_enhancement=True
)

print(processed.enhanced_text)

✨ What's New in v0.3.0

Simplified LLM Architecture

• Single comprehensive prompt: Handles all math conversion in one API call
• Professor-style narration: Natural explanations instead of robotic "subscript" language
• Intelligent chunking: Automatically handles large documents within OpenAI limits
• Better error handling: Clear failures instead of silent returns

Natural Mathematical Explanations

Before: $p_C$ → "p subscript C"

After: $p_C$ → "p underscore capital C, the proportion of compliers"

Complex expressions:

• $F_{1C}$ → "F underscore one capital C, the outcome distribution for treated compliers"
• $E = mc^2$ → "energy equals mass times the speed of light squared"

Key API Changes

• Main function: simple_llm_enhancer.enhance_document_simple()
• Smart chunking for documents > 128K tokens
• Single LLM call for most documents
• Professor-style math conversion prompt

Requirements

• Python 3.9+ (excluding 3.9.7)
• OpenAI API key (required for LLM enhancement)
• pydub (for audio chunking)
• PyPDF2 or PyMuPDF (for PDF processing)

Optional Dependencies

# For better PDF processing
pip install PyMuPDF

# For offline TTS
pip install pyttsx3

# For audio format conversion
# Install ffmpeg via your system package manager

Architecture

Paper Voice uses a clean modular pipeline:

PDF → LaTeX/Markdown → LLM Enhancement → TTS

1. PDF Extraction: Extract text with pdf_utils.extract_raw_text()
2. LLM Enhancement: Convert math to natural language with simple_llm_enhancer.enhance_document_simple()
3. Audio Generation: Create audio with tts.synthesize_speech_chunked()

Examples

Basic Usage

from paper_voice.simple_llm_enhancer import enhance_document_simple

# Simple math conversion
text = "The learning rate α controls convergence of $\\theta^* = \\arg\\min J(\\theta)$."
enhanced = enhance_document_simple(text, "your-api-key")
# Result: Natural professor-style explanation of the math

With Progress Tracking

def progress_callback(message):
    print(f"Progress: {message}")

enhanced = enhance_document_simple(
    content, 
    api_key, 
    progress_callback=progress_callback
)

Large Document Handling

The system automatically handles large documents:

• Documents < 128K tokens: Single LLM call
• Documents > 128K tokens: Intelligent chunking with natural breakpoints

Configuration

Set your OpenAI API key:

export OPENAI_API_KEY="your-key-here"

Or pass it directly to functions:

enhanced = enhance_document_simple(content, api_key="your-key")

License

MIT License - see LICENSE file for details.