复杂环境下语音增强的复平面谱减法  

Complex plane spectral subtraction for speech enhancement in complicated environments

作　　者：王水平[1,2,3] 唐振民[3] 陈北京[1,2] 蒋晔[3] 

机构地区：[1]南京信息工程大学计算机与软件学院,江苏南京210044 [2]南京信息工程大学江苏省网络监控工程中心,江苏南京210044 [3]南京理工大学计算机科学与工程学院,江苏南京210094

出　　处：《南京理工大学学报》2013年第6期857-862,共6页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(61103141);江苏省自然科学基金(CX2211_0261);江苏省高等学校大学生实践创新训练计划指导项目(301310300055)

摘　　要：为了降低复杂环境中存在的各类噪声,消除噪声给语音识别及说话人识别等系统在军事自动指挥及控制领域等的应用中带来的影响,设计了一种新的谱减法。针对传统谱减法在语音增强过程中的原始语音和噪声信号相差为0的错误假设,提出一种基于复平面的、考虑纯净语音与噪声及带噪语音之间相位差信息的改进谱减法,对增益函数进行修正。经过复杂的战场仿真环境测试表明,该方法性能较好,在低信噪比的环境下性能更优。对于输入信噪比为5 dB的带噪语音,经复平面谱减法去噪后,输出信噪比为12.21 dB,比传统谱减法提高了2.1 dB。To reduce noise in complicated environments and eliminate the inferences of noise on the applications of speech recognition and speaker recognition systems in automatic military command and control fields,a new spectral subtraction is designed. Aiming at the incorrect assumption of the traditional spectral subtraction that the phase difference between the clean speech and noise signals is zero in the speech enhancement process,an improved spectral subtraction is proposed based on complex plane and considering the phase difference among the clean speech,the noise and the noisyspeech signals. The gain function is corrected. The results of complicated battlefield simulation experiments show that the method proposed here performs better than the traditional spectral subtraction,especially in low signal noise ratio( SNR) conditions. For the noisy speech with input SNR of 5 dB,the output SNR is more than 12. 21 dB after denoised by using the complex plane spectral subtraction,which increases 2. 1 dB than the result of the traditional spectral subtraction.

关 键 词：复杂环境 语音增强 复平面 谱减法 增益函数 

分 类 号：N55[自然科学总论]