Summary of Denoising Lm: Pushing the Limits Of Error Correction Models For Speech Recognition, by Zijin Gu et al.
Denoising LM: Pushing the Limits of Error Correction Models for Speech Recognition
by Zijin Gu, Tatiana Likhomanenko, He Bai, Erik McDermott, Ronan Collobert, Navdeep Jaitly
First submitted to arxiv on: 24 May 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL); 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 The paper presents Denoising Language Model (DLM), an error correction model that significantly improves automatic speech recognition (ASR) performance by incorporating scaled training with synthetic data. The DLM is trained on vast amounts of text-to-speech (TTS) synthesized audio, which is then paired with original texts to correct ASR errors. The model’s key ingredients include up-scaling the model and data, using multi-speaker TTS systems, combining multiple noise augmentation strategies, and new decoding techniques. Experimental results show that DLM achieves state-of-the-art performance on Librispeech, matching self-supervised methods that use external audio data. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates a new language model called Denoising Language Model (DLM) to help with speech recognition. It’s like a proofreader for mistakes made by machines. The DLM is trained using lots of fake audio and text pairs, which helps it learn how to fix errors. The researchers found that the DLM works really well, even better than other methods that use extra audio data. This could lead to more accurate speech recognition in the future. |
Keywords
» Artificial intelligence » Language model » Self supervised » Synthetic data
