A Comparison between Deep Neural Nets and Kernel Acoustic Models for Speech Recognition

TL;DR

This paper compares large-scale kernel acoustic models and DNNs for speech recognition, finding that while kernel models match DNNs in some metrics, DNNs outperform in token-error-rates due to better posterior probability reduction, and introduces an entropy regularized perplexity technique to improve both models.

Contribution

The paper provides a comprehensive comparison between kernel methods and DNNs for acoustic modeling and introduces a novel entropy regularized perplexity technique for improved model selection and performance.

Findings

Kernel models match DNNs in perplexity and classification accuracy.

DNNs outperform kernel models in token-error-rates.

Entropy regularized perplexity improves recognition performance for both models.

Abstract

We study large-scale kernel methods for acoustic modeling and compare to DNNs on performance metrics related to both acoustic modeling and recognition. Measuring perplexity and frame-level classification accuracy, kernel-based acoustic models are as effective as their DNN counterparts. However, on token-error-rates DNN models can be significantly better. We have discovered that this might be attributed to DNN's unique strength in reducing both the perplexity and the entropy of the predicted posterior probabilities. Motivated by our findings, we propose a new technique, entropy regularized perplexity, for model selection. This technique can noticeably improve the recognition performance of both types of models, and reduces the gap between them. While effective on Broadcast News, this technique could be also applicable to other tasks.