摘要
为改善低信噪比环境下语音的质量,在传统MFCC特征提取的基础上,提出了两种提高识别系统鲁棒性的方法。一种方法利用混合窗函数对旁瓣的抑制来提高系统的鲁棒性;另一种方法是基于频谱峰值位置受背景噪声影响相对较小,将子带幅度信息和Mel子带频谱质心(MSSC)相结合。实验表明混合窗函数和子带频谱质心(MSSC)以及它们相结合的系统与使用传统MFCC的基准系统相比,在低信噪比的平稳噪声环境下系统的鲁棒性得到了一定的提高。
In order to improve the quality of speech in low SNR,two methods were proposed to improve the robustness of the system in this paper based on the traditional MFCC feature extraction.One is to use the side lobe suppression of mixed window function to improve the robustness of system;the other is to incorporate subband amplitude information with Mel-subband spectrum centroid(MSSC) because spectral peak position remains practically unaffected in the presence of background noise.Experimental results show that m...
出处
《计算机应用》
CSCD
北大核心
2009年第2期389-391,共3页
journal of Computer Applications
基金
湖南省科技计划项目(05FJ3046)
湖南省财政厅项目(湘财教指[2006]52号)