论文部分内容阅读
针对传统美尔倒谱系数特征提取方法不完全符合人耳听觉特性的问题,提出了一种基于伽马啁啾滤波器组的听觉特征提取算法.首先,给出了基于人耳耳蜗听觉模型的伽马啁啾滤波器组模型及其实现过程,并对其在频域的排列做了调整,再将输入语音与该滤波器组进行卷积,经过离散余弦变换得到伽马啁啾滤波器倒谱系数及其衍生特征.该算法模拟了人类听觉系统的信息处理机制,能准确表征出语音信号的特征,降低语音识别系统的难度.实验表明,与传统美尔倒谱系数相比,采用基于本文提出听觉特征的语音识别系统在识别率和鲁棒性上均有明显提高.
Aiming at the problem that the traditional method of Cepstral Coefficient Extraction does not fully accord with human auditory characteristics, a novel algorithm of auditory feature extraction based on gamma chirp filter is proposed.Firstly, based on the cochlear auditory model Gamma chirp filter bank model and its implementation process, and its frequency domain arrangement has been adjusted, and then the input speech and the filter bank convolution, after discrete cosine transform to get the gamma chirp filter down Spectral coefficients and their derivative features.The algorithm simulates the information processing mechanism of the human auditory system, can accurately characterize the characteristics of speech signals, reducing the difficulty of speech recognition system.Experiments show that, compared with the traditional Metric Cepstral Coefficients, In this paper, the speech recognition system with auditory features is obviously improved in recognition rate and robustness.