摘要
An objective method, using a multi-band analysis technique, was proposed for analyzing plosive consonants in cleft palate speech. At first, the speech signal is decomposed in frequency domain using an auditory filter-bank. Then, the sample-based features, namely cumulative energy and its increment speed, in each band were computed. Finally, using principle component analysis, these features were fused into one combined feature vector for assessment. Since the algorithm is based on perceptual properties of human auditory ear using non-uniform and multi-band analysis, the improvements of the consistence between the proposed approach and subjective evaluation are obtained.
An objective method, using a multi-band analysis technique, was proposed for analyzing plosive consonants in cleft palate speech. At first, the speech signal is decomposed in frequency domain using an auditory filter-bank. Then, the sample-based features, namely cumulative energy and its increment speed, in each band were computed. Finally, using principle component analysis, these features were fused into one combined feature vector for assessment. Since the algorithm is based on perceptual properties of human auditory ear using non-uniform and multi-band analysis, the improvements of the consistence between the proposed approach and subjective evaluation are obtained.
基金
supported by the National Natural Science Foundation of China(60875014,60772039)
