Research On Blind Separation Algorithm Of Indoor Reverberation Speech

In recent years,the blind separation of speech signals has gradually become a research hotspot in blind signal processing.It has broad market and application prospects in many fields such as mobile calls,speech recognition,speech positioning,and video calls.In a real indoor environment,there are multiple effects in the process of speech transmission from the source to the receiver,large reverberation components were caused by the delay,reflection,and refraction.Therefore,the signal received by a receiver is generally not linear and instantaneous mixing,but convolutional mixing.This paper has conducted an in-depth study on the theory of blind speech convolutional separation,and designed an experimental program for blind speech separation in a simulated indoor environment.The problem of residual crosstalk interference in the frequency-domain convolutional blind source separation algorithm and the problem of blind speech separation in the case of strong indoor reverberation are studied,and an improved algorithm is proposed to improve the separation performance of the algorithm.The main content of the paper is as follows:1.When the frequency domain Independent Component Analysis(ICA)algorithm performs short-time Fourier transform,its frame length must not only meet the requirements of the stationary time of the speech signal,but also be greater than the reverberation time.However,under normal circumstances,the two cannot be satisfied at the same time,the stationary time of the speech signal is much shorter than the reverberation time.Therefore,when the frame length is less than or almost equal to the reverberation time,the separation matrix calculated by the ICA algorithm will not completely converge,resulting in crosstalk interference in the separated speech signal and reducing the separation performance of the algorithm.To solve the above problems,this paper proposes a blind speech separation algorithm based on Wiener filtering and frequency domain ICA algorithm.First,the frequency domain ICA algorithm is used to separate the speech signal,and then the joint sorting algorithm is used to solve the ordering and amplitude uncertainty problems.Finally,Wiener filtering algorithm is used as a post filter at the end of the algorithm to filter out some interfering speech signals that may exist and residual noise,thereby improving and enhancing the separated speech signal.Finally,experimental simulations prove that the proposed algorithm significantly improves the separation effect.2.In a real indoor environment,especially in the case of strong reverberation,the performance of the blind speech separation algorithm will be severely degraded or even invalid.Under noise-free conditions,the quality of reverberant speech mainly depends on two different perceptual components: early reverberation and echo.This paper proposes a blind separation algorithm for indoor reverberation speech based on second-order dereverberation.The microphone receives a reverberated speech mixed signal,and first performs second-order dereverberation processing on it in two stages;in the first stage,an inverse filter is used to suppress early reverberation or increase the signal reverberation energy ratio;in the second stage,spectral subtraction is used to reduce the effect of echo.Then the target signal is transferred to the frequency domain,the independent vector analysis algorithm is used to avoid the ordering uncertainty problem,the speech signal is separated and finally restored to the time domain speech signal.However,when in a strong reverberation environment,the second-order dereverberation algorithm suppresses the degradation of the reverberation effect,resulting in crosstalk interference in the separated speech signal.For this reason,a blind separation algorithm for indoor reverberation speech based on Wiener filtering and second-order dereverberation is proposed.Experiments show that the proposed algorithm significantly improves the separation effect.