Skip to Main Content
A stress detection based on multi-class probabilistic support vector machines (MCP-SVMs) is proposed for classifying speech into following categories - no stress, primary stress, and secondary stress. The stress classifier is performed with a feature set including perceptual features, MFCC, delta-MFCC and delta-delta-MFCC. To observe that speakers from the same accent regions had similar tendencies in mispronunciations including word stress, this work uses English Across Taiwan (EAT) to represent Taiwanese-accented English speech corpora. The overall performance in the experimental results achieves about 84% classification of accuracy.