论文部分内容阅读
证实普通话可以分解为辅音音素和单元音音素通过过度音的连接,提出一种单字音特征提取方法。该方法在传统的帧特征提取基础上,对相关帧进行二次处理,得到单字语音中的多个代表帧,将代表帧进行拼接作为单字的特征矢量。这种特征提取方法能更好地表现说话人单字发音中相邻语音帧之间的连续性。仿真实验表明该方法在说话人识别系统的应用中达到较高的识别率,使识别时间进一步缩短。
It proves that Putonghua can be decomposed into consonant phonemes and monophone phonemes through the connection of excessive tone, and a method of extracting single-word features is proposed. Based on the traditional frame feature extraction, the method performs secondary processing on the related frames to obtain multiple representative frames in a single word speech, and splices the representative frames as a single-character feature vector. This feature extraction method can better represent the continuity between adjacent speech frames in speaker’s pronunciation. Simulation results show that this method achieves high recognition rate in speaker recognition system and shortens the recognition time.