摘要
在语音识别中,MFCC参数是说话人识别中常用的特征参数之一。文中针对说话人识别速度较慢以及占用资源较大的问题,提出了一种MFCC计算的有效方案。利用MFCC滤波器的频率响应函数的三角形结构,改进了Mel滤波器的设计方法。实验结果表明,文中所提方案在单帧内存访问时间上减少了83.6%,在保证识别准确率不降低的情况下,使识别速度大幅度提高,降低了说话人识别计算的复杂性。
The speaker recognition is an important research direction of speech recognition.The feature extraction is a key technology in the speaker recognition system.This paper proposes an efficient scheme for Mel Frequency Cepstral Coefficients(MFCC)computation.It exploits the triangular structure of the frequency response function of the MFCC filters.The simulation results show that the scheme proposed in this paper reduced the access time by 83.6%in single frame memory and the proposed scheme gained better latency with less memory.
作者
毕灶荣
童东兵
陈巧玉
BI Zaorong;TONG Dongbing;CHEN Qiaoyu(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science, Shanghai 201620,China;School of Statistics and Mathematics,Shanghai Lixin University of Accounting and Finance,Shanghai 201620,China)
出处
《电子科技》
2018年第4期25-28,共4页
Electronic Science and Technology
基金
国家自然科学基金(61673257
11501367)
中国博士后科学基金(2015M581528)
上海市自然科学基金(15ZR1419000)
