Summary of Mathspeech: Leveraging Small Lms For Accurate Conversion in Mathematical Speech-to-formula, by Sieun Hyeon et al.
MathSpeech: Leveraging Small LMs for Accurate Conversion in Mathematical Speech-to-Formula
by Sieun Hyeon, Kyudan Jung, Jaehee Won, Nam-Joon Kim, Hyun Gon Ryu, Hyuk-Jae Lee, Jaeyoung Do
First submitted to arxiv on: 20 Dec 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed MathSpeech pipeline integrates Automatic Speech Recognition (ASR) models with small Language Models (sLMs) to correct errors in mathematical expressions and accurately convert spoken expressions into structured LaTeX representations. This is particularly important for conveying mathematical concepts orally, as current ASR models often produce verbose and error-prone textual descriptions. The pipeline demonstrates LaTeX generation capabilities comparable to leading commercial Large Language Models (LLMs), while leveraging fine-tuned small language models of only 120M parameters. Evaluation on a new dataset derived from lecture recordings shows significant improvements in CER, BLEU, and ROUGE scores for LaTeX translation compared to GPT-4o. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary MathSpeech is a tool that helps people understand math better when it’s read out loud. Right now, computers can’t accurately turn spoken math into the correct written format, which makes it hard for people with hearing impairments or language barriers to follow along. The MathSpeech team created a new way to combine computer speech recognition with small language models to fix errors and get the right math equations. They tested this method on recordings of lectures and found that it works better than other big language models like GPT-4o. |
Keywords
» Artificial intelligence » Bleu » Cer » Gpt » Rouge » Translation
