【摘要】 基于超音段信息在语音感知中的显著作用。本文提出了一种新颖的汉语双音节词(二字词)识别方案。首先将输入语音调型进行时、频归一化处理,并将其和参考调型匹配;再对由此得到的候选集进行精确的谱匹配。在这步处理中结合了动态能量信息,并采用了修正的动态规划算法。实验结果表明,这种方案对于高混淆性汉语二字词识别十分有效。更多还原

【Abstract】 In this paper a novel approach is presented for the recognition of Mandarin bisyllabic words, which is based on the essential role the suprasegmentals play in speech perception. The pitch of the input utterance is first extracted, after undergoing time- and frequency- normalized procedure, it is then compared with a set of reference pitch contours, thus a search for the optimal candidates in the following fine matching is guided, in which a modified One-Stage Dynamic Programing algorithm is adopted by using a compound distortion metric of spectral and dynamic energy shape. The results show that the approach in question is particular suitable for recognizing highly confused Mandarin bisyllabic words.更多还原