多声源环境下的鲁棒说话人识别 robust speaker recognition in multi-source environment.pdfVIP

计 算 机 系 统 应 用 2015 年 第24 卷 第 4 期 多声源环境下的鲁棒说话人识别① 张凤仪, 夏秀渝, 冉国敬, 何 礼, 叶于林 ( 四川大学 电子信息学院, 成都 610065) 摘 要: 针对多声源干扰环境下说话人识别系统性能急剧下降的问题, 提出一种提取目标语音的前端处理方法, 该方法依据独立语音时频域的近似稀疏性, 基于目标语音方位信息采用非线性时频掩蔽方法提取目标语音. 建 立了基于梅尔倒谱系数(MFCC) 的高斯混合模型(GMM)说话人识别系统. 仿真实验证明, 该方法能有效提取目标 语音, 提高说话人识别系统的鲁棒性. 该文多声源干扰仿真实验条件下, 说话人识别系统的识别率平均提高了 25%左右. 关键词: 说话人识别; 语音增强; 方位信息; 时频掩蔽; MFCC 参数 Robust Speaker Recognition in Multi-Source Environment ZHANG Feng-Yi, XIA Xiu-Yu, RAN Guo-Jing, HE Li, YE Yu-Lin (College of Electronics and Information Engineering, Sichuan University, Chengdu 610064, China) Abstract: The Speaker Recognition System is significantly affected by the Multi-Sound sources problem. In order to overcome this problem, a target sound extraction algorithm named time-frequency masking is proposed. The proposed algorithm is based on the sound source azimuth information and the approximate sparse nature of sound. A Mel-frequency cepstral coefficient (MFCC) based Gaussian mixture model (GMM) speaker recognition system is presented to improve the recognition robustness. The proposed algorithm has been tested on the simulated data through a number of experiments which shows the efficiency and robustness of the proposed algorithm. In the Multi-Sound sources environment, the recognition rate of the proposed algorithm can be improved by about 25%. Key words: speaker recognition; speech enhancement; sound source azimuth information; time-frequency masking; MFCC 1 引言 数等. 前端语音增强方法有谱减法、维纳滤波法等. 当 随着计算机技术和信息化社会的发展, 说话人识 噪声近似为平稳宽带噪声时, 这两种方法是比较有效 别作为具有语音识别与理解功能的智能人T 机接口, 的. 其应用领域在不断扩大, 如保密安全、司法鉴定及军 然而, 实际