Study of Phonemes Confusions in Hierarchical Automatic Phoneme Recognition System

TL;DR

This paper analyzes phoneme confusions in a hierarchical recognition system, identifying pronunciation similarities that impact accuracy, and proposes a new system that improves recognition rates on the TIMIT database.

Contribution

It introduces a confusion-based hierarchical recognition approach that isolates problematic phonemes to enhance phoneme recognition performance.

Findings

Significant recognition rate improvements on TIMIT database

Confusions are mainly due to phoneme pronunciation similarities

Hierarchical recognizer outperforms previous models

Abstract

In this paper, we have analyzed the impact of confusions on the robustness of phoneme recognitions system. The confusions are detected at the pronunciation and the confusions matrices of the phoneme recognizer. The confusions show that some similarities between phonemes at the pronunciation affect significantly the recognition rates. This paper proposes to understand those confusions in order to improve the performance of the phoneme recognition system by isolating the problematic phonemes. Confusion analysis leads to build a new hierarchical recognizer using new phoneme distribution and the information from the confusion matrices. This new hierarchical phoneme recognition system shows significant improvements of the recognition rates on TIMIT database.