A cepstrum-based preprocessing algorithm for single-channel speech enhancement in adverse environments is proposed.The proposed preprocessing scheme can suppress the harmonic components of the voiced speech,and result in avoiding overestimating of the noise power spectral density (NPSD).Moreover,the preprocessing technique can also increase the number of speech gaps,in favor of noise estimation algorithms tracking the minimum of the signal energy.The experimental results reveal that the cepstrum-based preprocessing algorithm can track both the stationary and the non-stationary noise efficiently with much less overestimation of the NPSD,and achieve better performance in terms of both the segmental signal-to-noise ratio improvement (SegSNRI) and the perceptual evaluation of speech quality (PESQ).%提出一种应用于单通道语音增强技术的倒谱预处理算法.倒谱预处理通过抑制语音中浊音的谐频成分,避免在非平稳噪声跟踪过程中将语音浊音信号功率谱错误估计成噪声功率谱；同时预处理还能增加语音间隙出现的概率,跟踪搜索窗内信号能量最小值时更有效地跟踪到噪声能量电平.实验结果表明应用文中提出的预处理算法后,快速有效跟踪噪声的同时能避免噪声功率谱过估,并且在客观评价指标方面得到更高的段信噪比提高量(SegSNRI:Segmengtal Signal-to-Noise Ratio Improvement)和感知语音质量评价(PESQ:Perceptual Evaluation of Speech Quality)得分改善量.
展开▼
