G-Augment: Searching for the Meta-Structure of Data Augmentation Policies for ASR
Abstract
Data augmentation is a ubiquitous technique used to provide robustness to automatic speech recognition (ASR) training. However, even as so much of the ASR training process has become automated and more "end-to-end", the data augmentation policy (what augmentation functions to use, and how to apply them) remains hand-crafted. We present Graph-Augment, a technique to define the augmentation space as directed acyclic graphs (DAGs) and search over this space to optimize the augmentation policy itself. We show that given the same computational budget, policies produced by G-Augment are able to perform better than SpecAugment policies obtained by random search on fine-tuning tasks on CHiME-6 and AMI. G-Augment is also able to establish a new state-of-the-art ASR performance on the CHiME-6 evaluation set (30.7% WER). We further demonstrate that G-Augment policies show better transfer properties across warm-start to cold-start training and model size compared to random-searched SpecAugment policies.
Keywords
Cite
@article{arxiv.2210.10879,
title = {G-Augment: Searching for the Meta-Structure of Data Augmentation Policies for ASR},
author = {Gary Wang and Ekin D. Cubuk and Andrew Rosenberg and Shuyang Cheng and Ron J. Weiss and Bhuvana Ramabhadran and Pedro J. Moreno and Quoc V. Le and Daniel S. Park},
journal= {arXiv preprint arXiv:2210.10879},
year = {2022}
}
Comments
6 pages, accepted at SLT 2022. Updated with copyright