摘要
传统的DTW算法着重于时间规整和间距测量的概念,对数据的可靠性没有进行有效的分析,且对连续词的识别效果不明显。基于松弛起始点和分段思想的改进DTW算法,可以改善DTW算法的缺陷。通过对语音样本0～9在MATLAB6.5上的仿真实现与分析表明,采用改进后的DTW算法具有良好的语音识别效果。
The traditional DTW algorithm is mainly concerned with the concept of time-coordinate and space-measurement, but it can not provide effective analysis of data-reliability and distinct recognition for continuous words. An improved DTW algorithm that can overcome the limitations is presented based on the flabby initiative point and the subsection idea. Analysis of the simulation results from speech samples of zero to nine on the MATLAB6.5 shows that the improved DTW algorithm can provide better recognition of speech sounds.
出处
《福建工程学院学报》
CAS
2004年第2期149-151,175,共4页
Journal of Fujian University of Technology
关键词
语音识别
端点检测
线性预测
speech recognition
point examination
linear prediction
