DTW语音识别算法研究与分析

研究动态时间规整(Dynamic Time Warping)语音识别算法问题,传统动态时间规整方法需要存储较大的矩阵,直接计算将会占据较大的空间,计算量也比较大,对系统硬件要求比较高。为了减小DTW算法的运算量,提高识别速度,对DTW语音识别算法进行优化改进。将局部路径约束和整体路径约束相结合,仅在一个规定的宽度内搜索动态规划路径,计算累积匹配距离。仿真实验结果表明该方法不仅可以降低运算负载,提高识别速度,而且能在一定程度上提高语音识别率。

The Research and Analysis of DTW speech recognition algorithm

Speech recognition based on dynamic time warping algorithm is discussed in the paper. The traditional DTW dynamic searching method needs large matrix storing, large space, a great of calculation and higher hardware demanding. In order to reduce the computation of the DTW algorithm and improve the recognition speed, DTW speech recognition algorithm is optimized and improved. Considering the local path constraint and the overall path constraint at the same time, speech recognition is carried out under the limitation of search width. A large number of experiments show that the method can reduce the computational load, increase the recognition speed, but also to a certain extent, improve the speech recognition rate.