基于谱减法的装甲车辆语音降噪研究  

Research of speech denoising technology based on improved spectral subtraction algorithm in armored vehicle

作　　者：刘玉军[1] 孔洲[1] 徐万里[1] 张增[1] 杨新旺[1] 蒋冰舒 韩林呈[1] 

机构地区：[1]装甲兵工程学院信息工程系,北京100072 [2]总装备部后勤部,北京100101

出　　处：《计算机工程与设计》2010年第21期4657-4660,共4页Computer Engineering and Design

摘　　要：为降低装甲车辆内部强噪声对语音通信的影响,结合Mel算法、频率短时能量差、听觉掩蔽效应和经过改进的谱减法,提出了一种语音激活检测和语音增强方法,对传统谱减法噪声估计不精确和语音失真等问题进行了改进。该方法采用Mel频率对带噪语音进行语音激活检测,对噪声进行保守估计,替代一般谱减法采用的噪声统计均值。结合听觉掩蔽阈值对谱减法的相关系数进行动态调整,避免传统算法系数保持不变的不合理性。实验结果表明,该方法能很好抑制音乐噪声,提高带噪语音信噪比,改善语音的清晰度和可懂度。To reduce the effect of high interior noise in voiced communication inside the armored vehicle, a voice activity detection and speech enhancement method is presented, involving Mel, frequency short-term energy difference, auditory masking effect and an improved spectral subtraction. It is helpful to solve the problems resulted from the noise estimation imprecise and voice distortion caused by the traditional spectral subtraction. According to the method, Mel frequency is used in voice activity detection on the noisy voice to estimate the noise conservatively instead of the statistical mean value of the noise adopted by the general spectral subtraction. Auditory masking threshold is combined to dynamically adjust the coefficient related to the spectral subtraction, which avoids the irrationality caused by the unchanged coefficient in traditional algorithm. It is proved that the method can suppress music noise effectively, improve the speech-to-noise ratio of the noisy speech as well as the speech clarity and intelligibility.

关 键 词：谱相减 骨传导 语音激活检测 听觉掩蔽 Mel 

分 类 号：TN912.3[电子电信—通信与信息系统]