Saved in:
Bibliographic Details
Main Authors: Xie, Jiamin, Hansen, John H. L.
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2310.18450
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916797140172800
author Xie, Jiamin
Hansen, John H. L.
author_facet Xie, Jiamin
Hansen, John H. L.
contents In this paper, we present MixRep, a simple and effective data augmentation strategy based on mixup for low-resource ASR. MixRep interpolates the feature dimensions of hidden representations in the neural network that can be applied to both the acoustic feature input and the output of each layer, which generalizes the previous MixSpeech method. Further, we propose to combine the mixup with a regularization along the time axis of the input, which is shown as complementary. We apply MixRep to a Conformer encoder of an E2E LAS architecture trained with a joint CTC loss. We experiment on the WSJ dataset and subsets of the SWB dataset, covering reading and telephony conversational speech. Experimental results show that MixRep consistently outperforms other regularization methods for low-resource ASR. Compared to a strong SpecAugment baseline, MixRep achieves a +6.5\% and a +6.7\% relative WER reduction on the eval92 set and the Callhome part of the eval'2000 set.
format Preprint
id arxiv_https___arxiv_org_abs_2310_18450
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle MixRep: Hidden Representation Mixup for Low-Resource Speech Recognition
Xie, Jiamin
Hansen, John H. L.
Audio and Speech Processing
Artificial Intelligence
In this paper, we present MixRep, a simple and effective data augmentation strategy based on mixup for low-resource ASR. MixRep interpolates the feature dimensions of hidden representations in the neural network that can be applied to both the acoustic feature input and the output of each layer, which generalizes the previous MixSpeech method. Further, we propose to combine the mixup with a regularization along the time axis of the input, which is shown as complementary. We apply MixRep to a Conformer encoder of an E2E LAS architecture trained with a joint CTC loss. We experiment on the WSJ dataset and subsets of the SWB dataset, covering reading and telephony conversational speech. Experimental results show that MixRep consistently outperforms other regularization methods for low-resource ASR. Compared to a strong SpecAugment baseline, MixRep achieves a +6.5\% and a +6.7\% relative WER reduction on the eval92 set and the Callhome part of the eval'2000 set.
title MixRep: Hidden Representation Mixup for Low-Resource Speech Recognition
topic Audio and Speech Processing
Artificial Intelligence
url https://arxiv.org/abs/2310.18450