摘要
动态时间规整 DTW算法是一种运用动态规划原理计算时间矢量相似度的方法 ,在语音识别等领域有着广泛的应用。窗口法是针对 DTW算法的特点 ,采用了动态高效的空间管理办法 ,解决了通常的 DTW算法需要存储一个较大的矩阵 ,因而在存储空间有限的硬件系统中难以实现的问题 ,窗口法已经在实际系统中得到了应用。
It is difficult to implement DTW(Dynamic Time-Warping) in storage-limited hardware application because a large memory is needed to store a big searching matrix. We propose to use Shifting Window Method to implement DTW in a small RAM, which is a storage-limited hardware. Section 2 utilizes Figs.3 and 4 to explain the principles of such implementation and derive Eqs.(5), (6),and (7). At the end of section 2 we list the nine steps of our algorithm for such implementation. We applied our algorithm to implement DTW in a talking toy made by us. We built the system of this toy on ATMEL 90S8535 SC with a RAM whose memory storage is only 512 Bytes. The performance of this toy was very close to what was possible in experiments with normal DTW that had to be done on a PC with 64M RAM. The experiments show that our use of Shifting Window Method can indeed save much storage memory without sacrificing the effectiveness of searching.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2002年第4期540-543,共4页
Journal of Northwestern Polytechnical University
