MathReader: Text-to-Speech for Mathematical Documents

ICASSP 2025 (Oral)

This is the official repository for MathReader, an advanced TTS document reader for academic mathematical documents.

Demo page: https://hyeonsieun.github.io/MathReader_demo/

This paper was accepted for oral presentation at ICASSP 2025.

The experimental code and test dataset developed for our research can be found here.

---

How to use MathReader

1. Install Nougat and NVIDIA NeMo and transformers library in your development environment.

   • Nougat github

   • NVIDIA NeMo

   • You can also set up the environment using the following code through the mathreader_environment.yml file

     : conda env create -f ./mathreader_environment.yml

2. Create a folder named 'test_audio' in the same location as MathReader.py.

3. Modify line 102 in MathReader.py (Write the path of the PDF file you want to perform OCR on.).

4. Run python MathReader.py in the terminal.

Citation

@INPROCEEDINGS{HyeonICASSP25,
    author={Hyeon, Sieun and Jung, Kyudan and Kim, Nam-Joon and Ryu, Hyun Gon and Do, Jaeyoung},
    booktitle={ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
    title={MathReader : Text-to-Speech for Mathematical Documents}, 
    year={2025},
    pages={1-5},
    keywords={Text recognition;Error analysis;Pipelines;Optical character recognition;Graphics processing units;Signal processing;Real-time systems;Mathematical models;Text to speech;Speech processing;OCR;T5;TTS;document reader;LaTeX},
    doi={10.1109/ICASSP49660.2025.10890531}
  }