一种基于VQ的有限非特定人汉语短语识别方法

研究了基于VQ的有限非特定人汉语语音命令的识别方法，识别对象是有限的特定人群(如5到6人)、有限汉语短语。该文采用MFCC作为识别特征，利用改进的LBG算法训VQ码本，为了提高识别率和拒识率，提出了采用倒谱距离法的有效语音端点检测方法以及实用的拒识方法。实验结果证明，系统在具有背景噪声的一般办公环境下由有限人训练后，当训练的说话人与识别系统的距离在0.5m的范围内时，测试识别率达到99％以上，未训练说话人的拒识率达82％。

Speech Recognition of Mandarain Phrases Based on VQ for Limited Independent Speakers

A limited speaker, such as 5 or 6 persons, mandarin speech phrase recognition method based on VQ is given. In this paper, the MFCC features is used as recognition feature, and VQ codebook is trained by using improved LBG algorithm. In order to enhance recognition rate and rejection rate, this paper presents that Cepstral distance rule is adopted in speech endpoint detection, and effective rejection recognition rule, and recognition strategy. The experiment is proceeded under commonly environment with background noise. The experiment results demonstrate that more than 99% test recognition rate is achieved in this system when the distance between trained speakers and a recognition device is less than 0.5m, and no trained speakers rejection rate is 82%.