摘要
提出了一种基于仿生模式识别理论的非特定人连续语音关键词识别的新算法．该算法无需对待识别连续语音进行端点检测和分割，通过直接对特征提取后的连续语音进行动态搜索，得到待识别连续语音到各类关键词训练网络的距离随时间变化的曲线，通过动态搜索距离曲线上谷值的大小和数目来判断有多少关键词．通过对小词汇量、不同语速条件下的连续语音的测试，得到了良好的识别结果，验证了此方法的有效性．
Based on biomimetic pattern recognition theory, a novel speaker-independent continuous speech keyword-spotting algorithm is proposed. Without endpoint detection and division, a dynamic search is directly given to the feature-extracted continuous speech, and the minimum distance curve between continuous speech samples and every keyword-training net is got. Through investigating the vale-value and the numbers of the vales of the curve, the number of the keywords can be known. From tests of small vocabulary continuous speech with various speaking rate, better recognition results can be got, which proves the validity of the algorithm.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2006年第B07期1-5,共5页
Journal of Harbin Engineering University
关键词
仿生模式识别
连续语音识别
关键词识别
高维空间点覆盖
小词汇量
biomimetic pattern recognition
continuous speech recognition
keyword recognition
high dimensional space vertex covering
small vocabulary
作者简介
王守觉（1925-），男，教授，中国科学院院士．
