Summary of Large Language Models Are Efficient Learners Of Noise-robust Speech Recognition, by Yuchen Hu et al.
Large Language Models are Efficient Learners of Noise-Robust Speech Recognition
by Yuchen Hu, Chen Chen, Chao-Han Huck Yang, Ruizhe Li, Chao Zhang, Pin-Yu Chen, EnSiong Chng
First submitted to arxiv on: 19 Jan 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Sound (cs.SD); Audio and Speech Processing (eess.AS)
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 A recent breakthrough in large language models (LLMs) has led to generative error correction (GER) for automatic speech recognition (ASR), leveraging the linguistic knowledge and reasoning abilities of LLMs. This paper extends a benchmark dataset, HyPoradise, to noisy conditions and explores whether LLMs can be trained to denoise GER similar to robust ASR. The proposed approach extracts language-space noise embeddings from N-best lists to represent audio noise, promoting the denoising process. A knowledge distillation (KD) method is also designed to distill real noise information in audio embeddings into language-space noise. Experiments on various LLMs demonstrate a new breakthrough with up to 53.9% correction improvement in word error rate using limited training data. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models are helping computers understand speech better! They use special knowledge and thinking skills to correct mistakes. This paper makes it possible for these computers to work well even when the audio is noisy or hard to hear. The researchers did this by creating a new way to teach the computer about noise, using something called language-space embeddings. This helps the computer learn to remove noise from the speech, making it better understood. They tested their idea on several different models and saw big improvements in how well they could correct mistakes! |
Keywords
* Artificial intelligence * Knowledge distillation
