应用动态时间规整与矢量量化的语音识别算法

Speech recognition algorithm using dynamic time warping and vector quantization

提出了一种基于动态时间规整(DTW)的改进平均最小距离识别算法,改善了孤立词识别的鲁棒性并提高了识别率。同时对矢量量化(VQ)算法分析了不同码本大小下的识别率,并比较了各种算法的运算时间。通过在MatLab上实现特定人孤立词小词汇量语音识别,实验的结果表明:基于DTW算法的改进平均最小距离法识别率显著提高;码本较大时VQ算法的识别率最高;VQ算法的识别率一般高于DTW算法且运行时间短。

An improved mean minimum distance method based on DTW is proposed in this paper, it improves the robustness of isolated words recognition and increases the recognition rate. The recognition rate of VQ algorithm in different codebook size is also analyzed as well as the computing time of each algorithm. By realization of specific-person isolated-word smallvocabulary speech recognition on MatLab, the research shows that the rate of improved mean minimum distance method based on DTW improves remarkably and the rate of VQ algorithm is the highest in large codebook. VQ algorithm is usually higher than DTW in recognition rate and takes less computing time.