Research On The Algorithm Of Separation And Location Of Blind Source Signals In Strong Noise Environment

The human ear can automatically identify various types of sound sources in a noisy environment and focus on the specific sound of interest,while voice devices have limitations in this regard.Blind source separation is a good way to solve this problem.Blind source signal separation and localization technology is an important branch of speech signal processing.It can be seen everywhere in daily life and is widely used,such as speech recognition,telephone or video teleconferencing,hearing aids,and so on.The ultimate goal of this technology is to extract useful or specifically required sound sources,which can suppress other noises even in a strong noise environment,separate pure source signals and locate them.In terms of speech localization,positioning accuracy indicators have been continuously improved with the development of silicon microphone array technology in recent years.The main research content of this paper is the separation and location of blind source signals under the microphone array.For the separation and enhancement of blind source signals,traditional algorithms such as Wiener filtering or independent component analysis can solve this problem,but all have limitations.This paper combines the two methods and uses the FastICA algorithm based on negative entropy to feature extraction and separate of the speech signals.The Wiener filter algorithm is applied to minimize the mean square error between the estimated signal and the speech signal extracted from the features in the ICA domain.The pure speech signal is estimated from the received strong noise speech.The simulation results show that the algorithm's enhancement effect is nearly doubled compared with the previous algorithm.For the microphone array,the LMS adaptive beamforming method is derived.This algorithm has a poor speech enhancement effect in a low signal-to-noise environment.To address this problem,a new LMS algorithm based on wavelet packet threshold denoising preprocessing is proposed.The algorithm is verified by simulation and it can effectively suppress noise.Traditional adaptive beamforming algorithms rely on the performance of the blocking matrix,which reduces the noise suppression capability and system instability of the system.Based on the adaptive beamforming algorithm,this study uses the FastICA algorithm instead of the ANC module to extract the expected speech signals from the array received signals,which effectively improves the noise suppression capability and stability of the system.For signal localization,the generalized cross-correlation function method is analyzed.This method must assume that there is no correlation between signal and noise,but the above assumptions may not be satisfied in actual situations,and the cross-correlation method can only be used for integer multiples of the data sampling period.These factors will bring some errors to the delay estimation.For this application scenario,a novel method for delay calculation based on parameter identification is proposed.This method uses a finite impulse response filter(FIR)to simulate the signal delay process.A standard microphone sampled signal is used as an input,and another microphone sampled signal is used as an output.The delay time is obtained through the FIR parameter identification.The sound source position can be obtained based on the five-element cross microphone array.Finally,the circuit system environment is set up,including five-channel audio acquisition module,FPGA processing module,audio data processing module,SDK algorithm processing module,and output positioning display module.The algorithm is transplanted to the pynq-z2 development module.It can be seen that the speech enhancement effect is significantly improved and the positioning is also significantly accurate through experiments.