Enhancing dysarthric speech recognition through SepFormer and hierarchical attention network models with multistage transfer learning
R. Vinotha, D. Hepsiba, L. D. Vijay Anand, J. Andrew, R. Jennifer Eunice · 2024
Read Enhancing dysarthric speech recognition through SepFormer and hierarchical attention network models with multistage transfer learning by R. Vinotha, D. Hepsiba, L. D. Vijay Anand, J. Andrew, R. Jennifer Eunice (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-11603152.txt.
Description
Speech disorders make voices hard for computers to understand, and Vinotha and co-authors build a cleanup stage to help, in a 2024 study of dysarthric speech recognition. Their front end merges two ideas: an adversarial network that polishes damaged speech, and a separation transformer that untangles overlapping signals.
Behind the cleanup sit recognition models, transformers and conformers, some fitted with layered attention. Training starts on the large LibriSpeech collection, then adapts in stages to the speech of affected speakers. Without cleanup, the two base models read about sixty-nine and seventy percent of words correctly. Adding the attention design pushed scores to about seventy-one and seventy-three percent, with the enhancement stage helping further in the full results. The authors present the pipeline as a step toward tools that listen better.
Catalog record
| CREATOR | R. Vinotha, D. Hepsiba, L. D. Vijay Anand, J. Andrew, R. Jennifer Eunice |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 91 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |