基于改进语音特征提取方法的语音识别  

Speech Recognition Based on Improved Speech Feature Extraction

作　　者：李建文[1] 张晋平[1] 

机构地区：[1]陕西科技大学电气与信息工程学院,陕西西安710021

出　　处：《微电子学与计算机》2009年第7期230-233,共4页Microelectronics & Computer

基　　金：国家自然科学基金项目(60672001)

摘　　要：在分析语音特征提取方法基础上提出一种改进组合算法,并采用HMM声学模型和Viterbi算法进行模式训练和识别.实验结果表明,该算法在噪声环境中具有较好的鲁棒性,能有效提高噪声环境下中文连续语音识别的正确率,增强语音识别整体性能,因此在噪声环境下的语音识别系统中具有一定的实用价值.This paper analyses speech feature extraction and makes improvement of it, also adopts the HMM acoustic model and the Viterbi algorithm to do model training and recognition. The experimental results show that the method has good robustness in the noisy environment. It can effectively improve the accuracy of the continuous chinese speech recognition, enhance the overall performance of speech recognition, and has some practical value to the speech recognition system in the noisy environment.

关 键 词：语音识别 特征提取 最大似然线性转换 MEL频率倒谱系数 隐马尔柯夫模型 

分 类 号：TN912.34[电子电信—通信与信息系统]