Chain-based Discriminative Autoencoders for Speech Recognition

TL;DR

This paper introduces chain-based discriminative autoencoders (c-DcAE) and their hierarchical and parallel variants for speech recognition, improving robustness and accuracy by combining reconstruction and phonetic embedding objectives.

Contribution

It proposes new chain-based and hierarchical/parallel autoencoder models that enhance speech recognition by integrating mutual information and multi-task learning.

Findings

c-DcAE outperforms baseline systems on WSJ and Aurora-4 datasets.

Hierarchical and parallel structures improve robustness to noisy speech.

Mutual information-based objective enhances phonetic embedding quality.

Abstract

In our previous work, we proposed a discriminative autoencoder (DcAE) for speech recognition. DcAE combines two training schemes into one. First, since DcAE aims to learn encoder-decoder mappings, the squared error between the reconstructed speech and the input speech is minimized. Second, in the code layer, frame-based phonetic embeddings are obtained by minimizing the categorical cross-entropy between ground truth labels and predicted triphone-state scores. DcAE is developed based on the Kaldi toolkit by treating various TDNN models as encoders. In this paper, we further propose three new versions of DcAE. First, a new objective function that considers both categorical cross-entropy and mutual information between ground truth and predicted triphone-state sequences is used. The resulting DcAE is called a chain-based DcAE (c-DcAE). For application to robust speech recognition, we further extend c-DcAE to hierarchical and parallel structures, resulting in hc-DcAE and pc-DcAE. In these two models, both the error between the reconstructed noisy speech and the input noisy speech and the error between the enhanced speech and the reference clean speech are taken into the objective function. Experimental results on the WSJ and Aurora-4 corpora show that our DcAE models outperform baseline systems.