摘要
提出了一种利用线性预测残差去除语音中加性白噪声的方法 .含噪语音经过线性预测分析后 ,所得的线性预测残差可分为与语音相关的语音残差部分和与噪声相关的噪声残差部分 .当噪声为白噪声时 ,噪声残差能量与噪声能量的比值大于语音残差能量与语音能量的比值 ,并且随着含噪语音信噪比的增大 ,两者数值上的差距越来越大 .因此 ,从含噪语音中按一定比例 λ(0 <λ<1 )减去预测残差 ,即可抑制噪声残差能量 .这实际上抑制了含噪语音中的噪声能量 ,提高了其信噪比 ,从而达到在一定程度上去噪的目的 .同时 ,这样处理造成的语音能量损失是有限的 .特别是 ,上述方法中对语音成分的操作仅仅在预测残差内进行 ,对语音的共振峰影响很小 ,因而能够较好地保留语音信号的频谱结构 ,使语音品质不致降低 .实验结果表明 。
This paper proposed a new method on reducing additive white noise in speech signal using linear prediction residual. As the result of linear prediction analysis, the prediction residual of noisy speech can be divided into two parts: the speech residual related with speech and the noise residual related with noise. When the noise is white noise, the ratio of the energy of noise residual to that of noise is larger than the ratio of the energy of speech residual to that of speech. And larger the signal-noise-ratio of speech is, larger the difference between the former and the latter is. Based on it, a method on speech denoising was proposed that the noise residual is subtracted from the noisy speech signal in the proportion of λ(0<λ<1). By means of such process, the energy of noise residual will be reduced. And the energy of the noisy speech is also reduced and the signal-noise-ratio is heightened. Meanwhile, the loss of the energy of speech is limited. Specially, the operation to speech component is only taken in prediction residual in this method, which produces little impairment to the formant of the speech. So the spectrum structure of the speech is retained well, and the quality of the speech will not be reduced. The result of experiments indicates that the proposed method is simple and effective.
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
2002年第7期1021-1025,共5页
Journal of Shanghai Jiaotong University
