Dual-microphone Speech Enhancement Based On Differential Microphone Arrays

With the development of miniaturization of electronic devices,some conventional microphone arrays speech enhancement methods limited by the size of the arrays,the number of microphones,and the speed of operation are not suitable for it.The differential microphone arrays are small in size and have superior directionality.The balance between algorithm performance and applicability has been achieved and has received extensive attention and application.In this context,this dissertation studies the existing dual-microphone speech enhancement based on differential microphone arrays and makes some improvements to the algorithms.The details are as follow.First,the theoretical analysis and algorithm simulation of the dual-channel speech enhancement algorithm based on first-order differential microphone(FDM)arrays are performed.Through simulation experiments,it is found that the speech processed by this algorithm still has a certain amount of residual noise,and the speech intelligibility and signal-to-noise ratio are not ideal.The algorithm will cause a certain degree of damage to the speech and affect the quality of the speech,when the estimation of mute segment is inaccurate.Then,in order to reduce the residual noise and improve the speech intelligibility and signal-to-noise ratio(SNR),an improved algorithm combined with ideal ratio mask(IRM)is proposed.Firstly,the algorithm uses a first-order differential array to process the noisy speech signals collected by the two microphones to obtain speech channel signals and noise channel signals.The two-channel signals are filtered by a gammatone filter bank that can simulate the cochlear implant model to obtain their respective time-frequency representations.Then the ideal float mask for the speech channel signal is calculated.Finally,the enhanced speech signal is synthesized by using an ideal float mask.The simulation experiment results show that the proposed improved algorithm has a higher segmented signal-to-noise ratio compared with the reference algorithm and can obtain higher quality speech signals.Finally,the SNR estimation method of FDM-SS algorithm is improved.a SNR estimation algorithm based on differential arrays is proposed,which solves the problem of speech damage caused by the inaccurate SNR estimation of the original algorithm.The algorithm uses the two-channel signal processed by the differential array to construction equation to estimate the SNR of speech channel signals.Two different methods are given for the construction of the SNR solving equation.Then use the amplitude spectrum subtraction to eliminate the residual noise.The simulation experiment results show that the algorithm can effectively suppress the noise and improve the quality of speech.The denoising effect and the speech quality are better than the contrast algorithm.