家电控制系统的语音关键词识别算法研究

Research on Speech Keyword Spotting Algorithm for Control System of Household Electrical Appliances

介绍了一种适用于家电控制系统的语音关键词识别算法。算法包括对关键词的搜索与确认两部分。通过对关键词和非关键词分别建立模型,建立起一个并行识别网络。使用帧同步的Viterbi算法在网络上搜索关键词。为使关键词的识别结果精确,在对支持向量机(SVM)的原理进行简要分析后,提出将SVM作为分类器对基于HMM的最大似然搜索结果进行关键词确认。实验结果显示,联合采用HMM和SVM的关键词语音识别混合算法具有较高的性能。

A speech keyword spotting algorithm for the control system of household electrical appliances is introduced. Searching and verifying the given keywords are concluded in this algorithm. A parallel spotting network is established after modeling for the keyword and garbage separately. Then, frame-synchronous Viterbi arithmetic is used for searching keywords on the network. After a brief analysis for SVM（Support Vector Machine）, SVM is proposed to be a classifier to verify keywords which is found out by the HMM（Hidden Markov Model） based on ML（Maximum Likelihood） rule in order to obtain a precise spotting result. The experiment result shows that the hybrid HMM/SVM based on speech keyword spotting algorithm has a good performance.