论文部分内容阅读
为了提高语音识别系统对环境噪声的鲁棒性,本文通过对几种特征参数的噪声鲁棒性的研究,在感知倒谱分析(Mel-Cepstrum)的基础上,引入了多分辨率小波分析技术,利用小波变换的时频局域性并结合感知频域上的滤波技术,提出了一种新型的语音特征参数(DWT-MFC),该特征参数对于环境噪声具有很强的耐噪性。实验表明,与(LPCC)和(MFCC)参数相比,采用DWT-MFC参数使系统的噪声鲁棒性得到显著的提高。
In order to improve the robustness of speech recognition system to ambient noise, this paper introduces the noise robustness of several feature parameters based on the Mel-Cepstrum and introduces the multiresolution wavelet analysis , A new speech feature parameter (DWT-MFC) is proposed based on the time-frequency localization of wavelet transform combined with the filtering technology in the perceived frequency domain. The feature parameters are highly robust to environmental noises. Experiments show that compared with (LPCC) and (MFCC) parameters, the robustness of the system is significantly improved using DWT-MFC parameters.