非特定人问候语连续语音识别技术的研究  

The study on continuous speech recognition techqitue of greetings for speaker-independent

作　　者：叶虹[1] 王海伦[1] 姜春娣[1] 

机构地区：[1]浙江工业大学浙西分校,浙江衢州324000

出　　处：《浙江工业大学学报》2010年第5期566-569,共4页Journal of Zhejiang University of Technology

基　　金：浙江省自然科学基金资助项目(Y1090758)

摘　　要：为了解决连续的语音识别对端点的依赖性问题,提出了一种基于高维空间点覆盖动态搜索理论的非特定人问候语连续语音识别的新算法,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来判断识别结果.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.In order to solve the problem about the continuous speech recognition depending on the endpoint,a novel algorithm of continuous speech recognition of greetings for speaker-independent is presented.This method is designed based on high-dimensional space covering,dynamic search theory without endpoints detection and segmentation.This method can directly search the continuous speech after the features are extracted.The distance curve between continuous speech sample and each training net over time is gotten.The recognition result can be gotten by dynamically searching the distance to the valley value on the curve.The test with small vocabulary and continuous speech under various speaking rate have a good result.That means that this method is valid.

关 键 词：连续语音识别 高维空间点覆盖 非特定人 

分 类 号：TN912.3[电子电信—通信与信息系统]