Prosody Study with Context-Dependent Acoustic Models

Abstract

In this paper, we propose to study prosody with context-dependent acoustic models. We find that we can achieve better resolution on a specific aspect by training CDM with certain focus. For the tone recognition task, CDM with focus on tones should be used and it achieves 15.2% relative error reduction, when comparing with the traditional tri-phone models. For detecting prosody boundaries, CDM with focus on position should be used and the accuracy of prosodic word is 92.2%. CDMs are also used to visualize the f0 patterns of sentences with give contextual information. Such patterns are helpful to understand the interaction among contextual factors. Overall, CDMs are useful data source for various prosody studies.