噪声环境下MFCC特征提取

MFCC Feature Extraction in Noisy Environments

梅尔倒谱系数(MFCC)模拟了人耳的听觉特性,在语音识别实际应用中取得了较高的识别率。本文研究了在噪声环境下提取MFCC的一般过程和方法,研究了对噪声信号在时域与频域中的处理方法。最后用HTK工具箱进行实验验证文中所用方法的识别性能,本系统与基本特征提取方法相比,识别率有很大提高。

MFCCs （ Mel - Frequency-Cepstral Coefficients） is based on the human ears non-hnear frequency charactensnc and perform a high recognition rate in practical speech recognition application. In this paper, we introduced the general process and method of extracting MFCCs in noisy environment. The processes of noisy speech signal in time and frequency domain are proposed, For evaluating the improvement of speech recognition with the proposed methods, we used the HTK speech recognition toolkit to perform the experiments, the results showed that the performance was obviously improved.