MixSpeech: Data Augmentation for Low-resource Automatic Speech Recognition

TL;DR

MixSpeech introduces a mixup-based data augmentation technique for low-resource automatic speech recognition, improving accuracy over existing methods by combining speech features and recognition losses.

Contribution

The paper presents MixSpeech, a novel data augmentation method that enhances low-resource ASR performance by applying mixup to speech features and recognition losses.

Findings

MixSpeech outperforms baseline models without augmentation.

MixSpeech surpasses SpecAugment in accuracy, with a 10.6% PER improvement on TIMIT.

Achieves a WER of 4.7% on WSJ dataset.

Abstract

In this paper, we propose MixSpeech, a simple yet effective data augmentation method based on mixup for automatic speech recognition (ASR). MixSpeech trains an ASR model by taking a weighted combination of two different speech features (e.g., mel-spectrograms or MFCC) as the input, and recognizing both text sequences, where the two recognition losses use the same combination weight. We apply MixSpeech on two popular end-to-end speech recognition models including LAS (Listen, Attend and Spell) and Transformer, and conduct experiments on several low-resource datasets including TIMIT, WSJ, and HKUST. Experimental results show that MixSpeech achieves better accuracy than the baseline models without data augmentation, and outperforms a strong data augmentation method SpecAugment on these recognition tasks. Specifically, MixSpeech outperforms SpecAugment with a relative PER improvement of 10.6$\%$ on TIMIT dataset, and achieves a strong WER of 4.7$\%$ on WSJ dataset.