论文部分内容阅读
数码语音“２”与“８”等的混淆是汉语数码语音识别错误的主要来源。为此，提出了在汉语数码语音识别中引入声调判别的方法。首先设计了变长度短时平均幅度差函数（ＬＶＡＭＤＦ）、元音中心定位、基音谐波单周期校正以及基音邻近搜索等一系列高性能基音周期估计算法，在此基础上设计了一个针对汉语数码语音声调识别的ＭＤＴＤ算法。实验表明，新的基音周期估计方法和ＭＤＴＤ算法使汉语数码语音识别率由９５．２％上升到９８．５％，更使“２”与“８”的分辨率由９０．５％上升到了９８．８％，从而较好地解决了这个难题。
The confusion of digital voice “2” and “8” is the main source of Chinese digital voice recognition errors. For this reason, a method of introducing tone discrimination into Chinese digital speech recognition is proposed. Firstly, a series of high-performance pitch estimation algorithms such as LVAMDF, vowel center localization, pitch harmonic single-cycle correction and pitch-neighbor search are designed. Based on this, Speech Tone Recognition MDTD Algorithm. Experiments show that the new pitch period estimation method and MDTD algorithm increase the recognition rate of Chinese digital speech from 95.2% to 98.5%, and further increase the resolution of “2” and “8” from 90.5% to 98 .8%, so as to better solve this problem.