Automatic Detection of Stress in Mandarin Utterance with Tone Dependent Model

In this paper, we present the work in progress on automatic detection of stress in continuous Mandarin (standard Chinese) spoken utterance, and we are interested in finding the characteristic and performance of the acoustic stress cues in Mandarin. Therefore, correlated stress features including pitch, duration, intensity and spectral intensity are exploited with the purpose of developing the baseline system, and proper normalization is also implemented, we find that stress in Mandarin exhibits a strong correlation with duration and spectral intensity, while pitch and intensity are the secondary in importance. The combination of duration, spectral intensity and intensity parameters performs well with the average equal error rate (EER) of 10.29%. In respect that Mandarin is a tonal language, we propose a new method by using the tone dependent model (TDM) to alleviate the affection of tone in stress detection on the baseline system, experiment results show that the proposed method is feasible and provides respectable performance and the average EER gets to 7.07%, with a 3.22% reduction.