Orthogonal Gradient Penalty for Fast Training of Wasserstein GAN Based Multi-Task Autoencoder toward Robust Speech Recognition

Chao-Yuan KAO  Sangwook PARK  Alzahra BADI  David K. HAN  Hanseok KO  

Publication
IEICE TRANSACTIONS on Information and Systems   Vol.E103-D   No.5   pp.1195-1198
Publication Date: 2020/05/01
Online ISSN: 1745-1361
DOI: 10.1587/transinf.2019EDL8183
Type of Manuscript: LETTER
Category: Speech and Hearing
Keyword: 
speech enhancement,  generative adversarial networks,  deep learning,  robust speech recognition,  

Full Text: PDF(470.2KB)>>
Buy this Article




Summary: 
Performance in Automatic Speech Recognition (ASR) degrades dramatically in noisy environments. To alleviate this problem, a variety of deep networks based on convolutional neural networks and recurrent neural networks were proposed by applying L1 or L2 loss. In this Letter, we propose a new orthogonal gradient penalty (OGP) method for Wasserstein Generative Adversarial Networks (WGAN) applied to denoising and despeeching models. WGAN integrates a multi-task autoencoder which estimates not only speech features but also noise features from noisy speech. While achieving 14.1% improvement in Wasserstein distance convergence rate, the proposed OGP enhanced features are tested in ASR and achieve 9.7%, 8.6%, 6.2%, and 4.8% WER improvements over DDAE, MTAE, R-CED(CNN) and RNN models.