摘要
在语音识别系统中,表示语音信号的高维特征矢量会使系统复杂度上升。由于语音信号存在无用和冗余信息,为了去除冗余和不相关特征,当语音信号经过预处理、提取特征参数之后,使用蚁群优化算法在特征矢量空间中选择本质特征,从而以不牺牲识别率为前提完成降维。利用隐马尔科夫模型(HMM)进行语音识别的仿真结果表明,在汉语数字语音识别系统上,蚁群优化算法的性能优于传统的遗传算法和未进行特征选择的原始特征集合。分析结果显示利用蚁群优化的特征集合可以提高识别系统的性能,而且识别速度得到了提高。
In speech recognition system,high dimensionality feature vectors which represent speech signal can make complexity rise. In order to remove redundant and irrelevant features,a method based on ant colony optimization was employed after preprocessing and features extracting,which optimized dimensionality of feature space by selecting relevant underlying features. The simulations via a Hidden Markov Model in speech recognition show that the performance of the proposed algorithm is better than traditional genetic algorithm and feature set without optimization in Mandarin digit speech recognition system. The results of analysis indicate that with the optimized feature set,the performance of the recognition system is improved. Moreover,the speed of recognition is increased by using ant colony optimization.
出处
《计算机仿真》
CSCD
北大核心
2016年第2期409-412,417,共5页
Computer Simulation
基金
国家自然科学基金项目资助(61473041)
内蒙古高校科研项目(NJZY13139)
关键词
蚁群优化
特征提取
特征选择
语音识别
Ant colony optimization
Feature extraction
Feature selection
Speech recognition
