论文部分内容阅读
提出一种自适应帧长语音特征分析方法，使语音编码更准确，达到提高语音识别性能的目的。该方法包括过渡帧检测和过渡语音帧特征表示两方面。采用了两种特征表示方法。基于ＴＩＭＩＴ语音数据包和自定义的汉语语音数据的单词识别实验表明，这两种表示方法有相同的效果，都能在一定程度上提高识别系统的性能，但计算量稍有区别。基于ＴＩＭＩＴ数据的ＤＨＭＭ系统和ＣＨＭＭ系统的错识率分别下降了１１．２１％和９．５８％；基于自定义数据的ＤＨＭＭ系统和ＣＨＭＭ系统的错识率分别下降了１１．５５％９．５％。
A speech feature analysis method with adaptive frame length is proposed to make speech coding more accurate and achieve the purpose of improving speech recognition performance. The method includes two aspects: transitional frame detection and transitional speech frame feature representation. Two features are used. The experiments of word recognition based on TIMIT voice packets and custom Chinese speech data show that these two methods have the same effect, which can improve the performance of recognition system to some extent, but the computational complexity is slightly different. The misclassification rate of DHMM system and CHMM system based on TIMIT data decreased by 11.21% and 9.58% respectively; the error rate of DHMM system and CHMM system based on custom data decreased by 11.55% 9.5 %.