Pruning-aware Loss Functions for STOI-Optimized Pruned Recurrent Autoencoders for the Compression of the Stimulation Patterns of Cochlear Implants at Zero Delay
1 st Reemt Hinrichs, 2 nd Jörn Ostermann · 2025
Read Pruning-aware Loss Functions for STOI-Optimized Pruned Recurrent Autoencoders for the Compression of the Stimulation Patterns of Cochlear Implants at Zero Delay by 1 st Reemt Hinrichs, 2 nd Jörn Ostermann (2025) 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/arxiv-2502-02424.txt.
Description
Reemt Hinrichs and Jörn Ostermann shrink the neural nets that squeeze sound for cochlear implants, without losing speech clarity. Their 2025 paper deals with implants that stream audio wirelessly to a processor worn on the head. A deep autoencoder can cut the bit rate at no delay, which saves power, but past work ignored the size of the model itself.
Size matters in hearing gear, where compute is scarce. The team adds a loss term that feels the effect of pruning, the cutting of weak links, while the net still trains. Against the standard cut-by-size route, their way kept speech scores far higher at steep cut rates. After fine-tuning, scores held near flat up to cuts of about half the model.
Catalog record
| CREATOR | 1 st Reemt Hinrichs, 2 nd Jörn Ostermann |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-02424