一种新的可应用于声目标识别的倒谱系数

摘要

A kind of cepstrum coefficients based on new feature Was presented.Compared with Melfrequency cepstral coefficients(MFCC) used extensively at present,it was improved for computing the new cepstrum feature in acoustic target recognition that 1)the bank of trianguiar filters Was improved based on cochlear model;2)the log-compressed Was replaced by the exp-compressed which is the function of frequency;3)the adaptive mechanism was introduced into the process of transformation from frequency domain to perceptual domain.Simulations of the four acoustic target recognition were carried out.The simulated results show that the proposed features have strong anti-noise ability.%提出一种新的倒谱系数,与目前广泛采用的美尔倒谱系数(MFCC)相比有以下改进:1)基于听觉外周模型改进了美尔频率倒谱系数(MFCC)的三角滤波器;2)用与频率相关的指数压缩代替固定的对数压缩;3)分析了频率的临界带变换在语音识别中所起的作用,从声目标识别的角度提出在临界带变换中引入与信号相关的自适应机制.通过4组声目标识别仿真对比试验,表明了新的倒谱系数比MFCC在抗噪声能力方面有明显提升.