摘要
根据MFCC中DCT的特点,设计一种基于DA算法的实现结构,采用先分解ROM再偏移二进制编码的方法对DA算法进行优化,将ROM表的大小由2~N减小到(N/K)2^(K-1)。通过仿真与FPGA测试,验证了该设计的正确性,能够满足说话人识别中MFCC参数提取的实时性要求和精度要求。
This paper presents an implementation structure based on Distributed Arithmetic（DA） according to DCT character in MFCC, which optimizes DA by using ROM reduction and offset binary coder, and reduces the size of ROM table from 2^N to （N/K）2^k-1. The results of simulation and FPGA test show this kind of design is correct, which meets the requirement of real-time and precision in MFCC computation for speaker recognition.
出处
《计算机工程》
CAS
CSCD
北大核心
2009年第5期265-267,共3页
Computer Engineering
基金
国家"863"计划基金资助项目(2006AA01Z425)
关键词
说话人识别
美尔频率倒谱系数
离散余弦变换
分布式算法
speaker recognition
Mel-Frequency Cepstral Coefficients（MFCC）
discrete cosine transform
distributed arithmetic
