Increasing demand for on-device Automatic Speech Recognition (ASR) systems has resulted in renewed interests in developing automatic model compression techniques. Past research have shown that AutoML-based Low Rank Factorization (LRF) technique, when applied to an end-to-end Encoder-Attention-Decoder style ASR model, can achieve a speedup of up to 3.7x, outperforming laborious manual rank-selection approaches. However, we show that current AutoML-based search techniques only work up to a certain compression level, beyond which they fail to produce compressed models with acceptable word error rates (WER). In this work, we propose an iterative AutoML-based LRF approach that achieves over 5x compression without degrading the WER, thereby advancing the state-of-the-art in ASR compression.
@article{arxiv.2008.02897,
title = {Iterative Compression of End-to-End ASR Model using AutoML},
author = {Abhinav Mehrotra and Łukasz Dudziak and Jinsu Yeo and Young-yoon Lee and Ravichander Vipperla and Mohamed S. Abdelfattah and Sourav Bhattacharya and Samin Ishtiaq and Alberto Gil C. P. Ramos and SangJeong Lee and Daehyun Kim and Nicholas D. Lane},
journal= {arXiv preprint arXiv:2008.02897},
year = {2020}
}