基于HTK的白族语音识别方法  

The Speech Recognition Method of the Bai's Language Based on HTK

作　　者：张令通[1] 

机构地区：[1]大理学院工程学院,云南大理671003

出　　处：《大理学院学报（综合版）》2013年第10期27-32,共6页Journal of Dali University

基　　金：云南省教育厅科学基金资助项目(2012Y154)

摘　　要：利用计算机识别少数民族语音是保护和传承民族文化的重要手段。白族是祖国西南边陲重要的少数民族之一,其历史悠久,文化灿烂。为实现使用白族语进行人与计算机的语音交互,提出了一种基于HTK的白族语音词识别方法。该方法针对白族语的发音特点,以音素为基本识别单元,利用HTK工具提取39维MFCC语音特征参数,构建HMM模型,采用Viterbi算法进行模型训练和匹配来实现白族语音的识别。实验表明,算法的识别准确率达到93.3%。该方法识别准确率高,为研究少数民族语音识别提供了有益的借鉴。Using computer to identify the speech sounds of minority nationality is an important means to protect and inherit the nation- al culture. The Bai is one of the important minority nationalities in the southwest border of the motherland. It has a long history and the splendid culture. In order to implement human-computer interaction in Bai＂ s language, a sound recognition method for Bai＇ s language based on HTK is proposed. According to characteristics of Bai＇s language, this method structures the HMM model through using pho- nemes as the basic units and using the HTK tool to extract the speech feature parameters of 39-dimensional MFCC. Viterbi algorithm is used to train the model and match to achieve speech recognition of Bai＇s language. The experimental result shows that the recogui- tion system can obtain a high recognition rate （up to 93.3% ） and also provides a useful reference for the speech recognition research on minorities＂ languages.

关 键 词：隐马尔可夫模型工具箱 白族语 识别 MEL频率倒谱系数 音素 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]