M.-T. Lin (Taiwan)
Endpoint Detection、Sub-Syllabic Units、Consonant/Vowel Boundary Detection 、 Coincidence Rate of Difference Margin.
A new system for the automatic segmentation and labeling of Mandarin speech is presented. The system is capable of labeling speech generated without requiring extensive linguistic knowledge or large training databases. In this work, a new approach using likelihood ratio test was used to design a speech/nonspeech classification and consonant/vowel segmentation method with a high accuracy rate for Mandarin sub-syllable recognition systems. The proposed method consists of feature extraction of the critical-band scaled short-time discrete Fourier transforms (CB-STDFT) followed by a decision rule based on the likelihood ratio test for the boundary detection. Our experiment results showed that the spectral feature STDFT has high noise immunity, the perceptual weighting via the critical-band frequency mapping can improve the accuracy of syllable boundary detection, and the likelihood ratio test also outperforms other threshold methods.
Important Links:
Go Back
