听觉频域掩蔽效应的自适应β阶贝叶斯感知估计语音增强算法

Adaptiveβ-order perceptually motivated speech enhancement algorithm based on frequency-domain auditory masking

为了在噪声抑制和语音失真中之间寻找最佳平衡,提出了一种听觉频域掩蔽效应的自适应β阶贝叶斯感知估计语音增强算法,以期提高语音增强的综合性能。算法利用了人耳的听觉掩蔽效应,根据计算得到的频域掩蔽阈自适应调整β阶贝叶斯感知估计语音增强算法中的β值,从而仅将噪声抑制在掩蔽阈之下,保留较多的语音信息,降低语音失真。并分别用客观和主观评价方式,对所提出的算法的性能进行了评估,并与原来基于信噪比的自适应β阶贝叶斯感知估计语音增强算法进行了比较。结果表明,频域掩蔽的β阶贝叶斯感知估计方法的综合客观评价结果在信噪比为-10 dB至5 dB之间时均高于基于信噪比的自适应β阶贝叶斯感知估计语音增强算法。主观评价结果也表明频域掩蔽的β阶贝叶斯感知估计方法能在尽量保留语音信息的同时,较好的抑制背景噪声。

Aimed to reach a compromise between noise suppression and speech distortion,an adaptiveβ-order perceptually motivated speech enhancement algorithm based on frequency-domain auditory masking to improve speech enhancement performance is proposed.The algorithm applies auditory masking effect.Theβvalue in the algorithm is adjusted adaptively according to the calculated frequency-domain masking threshold so that the noise is suppressed just under the threshold.In this way,the algorithm could maintain more speech information and reduce speech distortion. The performance of the algorithm was evaluated by subjective and objective measurements respectively,and was compared to that of adaptiveβ-order perceptually motivated speech enhancement algorithm based on signal to noise ratio. Results showed that when the SNR is between—10 dB and 5 dB the objective measurement results of the adaptiveβ-order perceptually motivated speech enhancement algorithm based on frequency-domain auditory masking is higher than that of the comparing algorithm.Subjective measurement results also showed that the adaptiveβ-order perceptually motivated speech enhancement algorithm based on frequency-domain auditory masking can suppress background noise well while saving speech information as much as possible.