Comparing human and machine vowel classification

Abstract

In this study we compare human ability to identify vowels with a machine learning approach. A perception experiment for 14 Hungarian vowels in isolation and embedded in a carrier word was accomplished, and a C4.5 decision tree was trained on the same material. A comparison between the identification results of the subjects and the classifier showed that in three of four conditions (isolated vowel quantity and identity, embedded vowel identity) the performance of the classifier was superior and in one condition (embedded vowel quantity) equal to the subjects’ performance. This outcome can be explained by perceptual limits of the subjects and by stimulus properties. The classifier’s performance was significantly weakened by replacing the continuous spectral information by binary 3-Bark thresholds as proposed in phonetic literature [8]. Parts of the resulting decision trees can be interpreted phonetically, which could qualify this classifier as a tool for phonetic research.