摘要
语音控制作为一种新型的人机交互手段,给用户带来更多的操作体验,在很多特定场景中具有必要性。本文将梅尔倒谱系数(MFCC)作为语音特征参数,采用动态时间规整算法(DTW)进行模式识别和分类,实现了小样本孤立词汇的实时识别,具有高识别率。在基本算法的基础上进行了边界条件改进,克服了端点检测缺陷。在语音特征提取上,分析比较了线性预测系数(LPC)和梅尔倒谱系数(MFCC)作为特征参数的优缺点,最后选定基于人耳听觉特性的MFCC作为语音特征参数。语音信号采用NI公司USB-6218采集卡将数据直接传输至MATLAB开发平台,在MATLAB集成环境下实现了语音识别程序。实验结果表明,系统可以实现6个特定的孤立词识别,满足实时性和准确性要求。
Speech control,as a new type of human-computer interaction method,brings better operation experience to users,and it is necessary in many specific scenes.In this paper,the MFCC are used as speech feature parameters,and the dynamic time warping algorithm(DTW)is used for pattern recognition and classification,which realizes the real-time recognition of small sample isolated words with high recognition rate.On the basis of the basic algorithm,the boundary condition is improved and the defect of endpoint detection is overcome.In the speech feature extraction,analysis and comparison of the linear prediction coefficient(LPC)and MFCC advantages and disadvantages as characteristic parameters,finally selected based on human auditory characteristics MFCC as speech feature parameters.The voice signals are directly transmitted to the MATLAB development platform by NI company's USB-6218acquisition card,and the voice recognition program is implemented in the MATLAB integrated environment.Experimental results show that the system can implement6specific isolated word recognition,which meets the requirements of real-time and accuracy.
出处
《科技资讯》
2017年第26期28-31,共4页
Science & Technology Information
基金
重庆市高等职业技术院校新技术推广项目(项目编号:GZTG201606)
第三批重庆市高等学校青年骨干教师资助计划(2016年11月发布)
