Summary of Token-weighted Rnn-t For Learning From Flawed Data, by Gil Keren et al.
Token-Weighted RNN-T for Learning from Flawed Data
by Gil Keren, Wei Zhou, Ozlem Kalinli
First submitted to arxiv on: 26 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: 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 The paper proposes a novel approach to training Automatic Speech Recognition (ASR) models using a token-weighted Recurrent Neural Network-Transcription (RNN-T) criterion. The traditional cross-entropy method optimizes the probability of all tokens in the target sequence, but this can lead to accuracy loss due to transcription errors. The new objective uses token-specific weights to de-emphasize error-prone tokens. This approach is particularly useful for semi-supervised learning with pseudo-labels and mitigates accuracy losses caused by human annotation errors. Experimental results show a consistent accuracy improvement of up to 38% relative using the token-weighted RNN-T method. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Researchers are working on making speech recognition computers better at understanding what people say. They want to make sure that mistakes in the data they use to train these computers don’t affect how well they work. To do this, they’ve come up with a new way of training these computers using something called token-weighted RNN-T. This new method helps the computers ignore mistakes and focus on what’s important. It works particularly well when there are mistakes in the data that was used to train the computer. The results show that this new method can make the computers more accurate, with improvements of up to 38%. |
Keywords
» Artificial intelligence » Cross entropy » Neural network » Probability » Rnn » Semi supervised » Token
