Speech-to-LaTeX: New Models and Datasets for Converting Spoken Equations and Sentences

Summary

This paper talks about Speech-to-LaTeX, a new dataset and AI models that convert spoken math expressions and sentences into LaTeX, which is a special code used to write math formulas clearly. These models work better than previous ones and support multiple languages.

What's the problem?

The problem is that math expressions spoken out loud are hard for AI to understand and turn into LaTeX accurately, especially when dealing with complicated equations or different languages. Earlier systems often made mistakes or couldn't handle both math and regular sentences well.

What's the solution?

The paper introduces a new open-source dataset with many examples of spoken math and sentences paired with their correct LaTeX versions. It also presents new models trained on this data to better recognize and convert spoken math expressions into LaTeX code with higher accuracy, and they can handle multiple languages effectively.

Why it matters?

This matters because being able to turn spoken math into clear LaTeX code makes it easier for students, teachers, and researchers to create and share math content quickly. It helps people who find typing math hard and supports learning and communication in different languages, making math more accessible to many.