摘要
在说话人识别方面,最常用到的语音特征就是梅尔倒频谱系数(MFCC)。提出了一种改进的提取MFCC参数的方法,对传统的提取MFCC过程中计算FFT这一步骤进行频谱重构,对频谱进行噪声补偿重建,使之具有很好的抗噪性,逼近纯净语音的频谱。实验表明基于此改进提取的MFCC参数,可以明显提高说话人识别系统的识别率,尤其在低信噪比的环境下,效果明显。
In the speaker recognition, Mel Frequency Cepstrum Coefficient(MFCC)is the most commonly used speech features. This paper presents an improved method of extraction to take the MFCC parameters, in the FFT of this step in the traditional process of extraction of MFCC spectrum reconstruction, noise compensation for reconstruction of the spectrum, with good noise immunity, approaching pure voice spectrum. The experiments show that the improvements based on this extracted MFCC, can significantly improve the recognition rate for speaker recognition system, especially in low SNR environment, the effect is obvious.
出处
《计算机工程与应用》
CSCD
2014年第7期217-220,共4页
Computer Engineering and Applications
基金
江苏省自然科学基金(No.BK2010546)
关键词
MFCC参数
频谱重建
说话人识别
MFCC parameters
spectrum reconstruction
speaker recognition