基于禁止搜索的非线性时间匹配优化算法

动态时间规整算法DTW(Dynamic Time Warping)作为一种非线性时间匹配技术已成功地应用于语音识别系统中。DTW算法使用动态规划技术来搜索两个时间序列的最优规整路径,虽然这种算法计算量小,运算时间较短,但只是一种局部优化算法。禁止搜索TS(Tabu Search)算法是一种具有短期记忆的广义启发式全局搜索技术,适用于解决许多非线性优化问题。本文将该技术用于语音识别系统中,提出了基于禁止搜索的非线性时间规整的优化算法TSTW,使得时间规整函数尽可能逼近全局最优。仿真结果表明,TSTW比DTW有更高的识别率,且运行时间比遗传时间规整算法GTW大大减少。

An algorithm for optimizing the nonlinear time alignment based on tabu approach

Dynamic Time Warping(DTW) has been widely used in speech recognition sys-tems as a, nonlinear time alignment technique. It uses the dynamic programniing technique to search the optimal warping path for two time sequences. Although this algorithm needs less computation and shorter training and searching time, it, is a local optimization algorithm. The Tabu Search(TS) algorithm is the generalized heuristic global search tedmique with short-time memory. and suitable for solving many nonlinear optimization problems. This paper applies this teclmique to speech recognition systems, and presents a new algorithm for optimizing time warping based on TS approach, which makes time warping functions optimized globally. Sim-ulation results show that TSTW has better time warping performance than DTW and GTW.