Chinese Accent Detection Using Acoustic Feature Sets with Context Features

Abstract

Accent is a critically important component of spoken communication. In this paper, we use acoustic features with context features to detect Chinese accent, and use NaiveBayes classifier to conduct model based on ASCCD reading discourse corpus. For comparison purposes, we digitally added white noise to ASCCD, and regarded it as ASCCD (white). NaiveBayes classifier is utilized to model the acoustic feature sets, which adequately present the property of the current syllable and context. The experimental results indicate that the acoustic feature sets with context features achieve 80.7% accent detection accuracy on ASCCD. The experiment also demonstrate that the acoustic feature sets with context features achieve 75.7% accent detection accuracy on ASCCD(white). Across these acoustic feature sets, we find pitch features to be the most predictive of Chinese accent on ASCCD and on ASCCD (white).Context is important to Chinese accent detection on ASCCD and on ASCCD (white).And the experimental results show that the NaiveBayes classification has a good classification result for these.