【摘要】 提出了一种基于听觉掩蔽效应的语音增强算法。算法对含噪语音的增强包括谱减法语音增强和感知加权语音增强两个步骤,分别从客观和主观两方面来提高含噪语音的质量。在谱减法阶段考虑了语音和噪声谱的交叉项,有效地减少了增强语音中的残余噪声;在感知加权处理中充分利用了人耳的掩蔽效应,设计了感知加权滤波器,对谱减法增强后的语音进行滤波,进一步消除残余噪声。对算法进行了数值实验,实验结果表明该算法能有效提高含噪语音的质量,比传统的谱减法有较大改进。更多还原

【Abstract】 A speech enhancement algorithm based on auditory masking effect is presented. The algorithm has two steps for the enhancement of noisy speech, which are spectral subtraction enhancement and perceptual weighting enhancement. The quality of the noisy speech from the objective and subjective aspect is improved separately. In spectral subtraction, the contribution of the cross-terms of speech and noise spectrum is taken into account. Therefore, the residual noise in the enhanced speech is reduced effectively. And in perceptual weighting stage, auditory masking effect is used. A perceptual weighting filter is designed to filter the enhanced speech by spectral subtraction so as to remove residual noise. Finally,numeric experimental results show the quality of noisy speech is improved effectively by the proposed algorithm which is better than the traditional spectral subtraction.更多还原