一种机器人智能语音识别算法研究

发布时间：2018-02-28 18:46

本文关键词： 端点检测特征提取 HMM算法语音识别　出处：《计算机测量与控制》2014年10期 　论文类型：期刊论文

【摘要】：针对智能机器人在非特定人语音识别中识别率偏低的问题,提出了一种双门限的端点检测算法,精确地检测出了语音端点,对分形维数和Mel频率倒谱系数(MFCC)进行结合,同时基于隐马尔可夫(HMM)模型,提出了智能机器人命令识别系统;在实验室环境下,利用Cool Edit软件录制了5男5女的语音,采样率为8kHz,精度为16位,内容为5个命令词,每个词均被采集6次,将每人的前3次发音作为模板语音,后3次发音作为测试语音,实验结果表明,系统识别率可以达到85%以上,MFCC与分形维数混合的语音特征参数的算法提高了系统识别率,优化了系统性能;该方法用于非特定人语音智能识别是可行的、有效的。
[Abstract]:In order to solve the problem of low recognition rate of intelligent robot in independent speech recognition, a two-threshold endpoint detection algorithm is proposed to detect the speech endpoint accurately. The fractal dimension is combined with the Mel frequency cepstrum coefficient (MFCC). At the same time, based on the hidden Markov model, an intelligent robot command recognition system is proposed. In the laboratory environment, the speech of 5 men and 5 women are recorded by Cool Edit software. The sampling rate is 8 kHz, the precision is 16 bits, and the content is 5 command words. Each word was collected for 6 times. The first three pronounciations of each person were used as template pronunciation and the latter three times as test pronunciation. The experimental results showed that, The algorithm of speech feature parameters mixed with fractal dimension and system recognition rate can reach more than 85%. The algorithm improves the recognition rate and optimizes the system performance, and it is feasible and effective to apply this method to the intelligent recognition of non-specific speech.
【作者单位】：重庆电子工程职业学院计算机学院;黄淮学院动画学院;
【基金】：重庆市教委“科学技术研究项目”(KJ132207) 重庆市教育科学院2013年度“十二五”规划课题(2013-ZJ-079)
【分类号】：TP242;TN912.34

【参考文献】