基于CHMM语音识别特征参数的选择方法

Selection Method of Feature Parameters for CHMM-Based Speech Recognition

基于CHMM的语音识别系统识别率高,但却占用系统资源较大,从而限制了其在资源受限的实际应用环境的有效实现。针对上述问题,给出特征参数选择的理论依据,弥补以往研究仅从实验结果分析,缺少理论依据的不足;同时提出根据各特征参数对系统误识率的影响程度来选择特征参数的新方法。该方法能使系统在训练,识别过程中的计算量和存储量明显减小,同时系统误识率不会显著改变。这为资源受限的语音识别系统,提供新的思路和有效的特征参数选择方法。

Speech recognition system based on CHMM has higher recognition rate but costs more system resources, which limit it＇s used in practice effectively when resources of system is limited. Firstly, the paper discussed theoretical foundation （or the selection of feature parameters for CHMM - based Speech recognition system, which supplies a gap for prevenient researches that only gave experiments without theory. Secondly, a new method was presented, according to the influence of the components of feature parameters on error rate, which can reduce the computation and storage in training and recognition, but has little effect on error rate of the system. The method offers more practical parameter combination for speech recognition system when its resource is limited