Research On Speech Enhancement Algorithms Based On Microphone Array And Implementation Of TMS320C6678

In the actual environment,noise is always everywhere,and speech signal is inevitably affected by various kinds of interference noise.The existence of these noises will not only affect the communication between people,but also reduce the performance of intelligent speakers,digital hearing aids and other electronic products depending on voice input.Therefore,the enhancement of noisy speech signal is not only necessary,but also has a broad application prospect.In this paper,according to the requirements of the actual project,studied the traditional microphone array speech enhancement algorithm and improved it on the basis of the microphone linear arra y composed of four microphones.Built a voice noise reduction prototype system which can be used in real environment,and the improved algorithm is successfully transplanted to the system and tested.The main research contents and results are as follows:Firstly,it introduced the research background and significance of the subject.Compared with the traditional single-microphone speech enhancement method,it introduced the advantages of using microphon e array for speech enhancement.The application scenarios of the microphone array and the research status of the microphone array at home and abroad are briefly described.Secondly,it described some basic knowledge of speech signal processing,such as pre-filtering,windowing,framing,etc.,then introduced the characteristics of the noise and reverberation of the model of the microphone array processing speech signal.Finally,several speech enhancements' performance criterions are introduced in detail.Thirdly,the basic principles of LCMV adaptive beamforming and GSC two traditional microphone array speech enh ancement methods are expounded.In view of the insufficiency of the accurate time delay estimat ion for the GSC method,had introduced the TF-GSC method which using the transfer function ratio.Although the TF-GSC method does not require accurate time delay estimation,the method of using the transfer function ratio to estimate the fixed beamformer will cause errors,and the TF-GSC method based on GSC structure has a weak ability to s uppress incoherent noise.Therefore,this article is optimized it from the following two aspects.First,a scaling factor K is introduced to optimize the fixed beamformer.The simulation results show that the output segmentation SNR of the improved method is higher than the previous method.Second,it is optimized by a psychoacousticbased post filtering method.And the simulation results show that the post-filtering method can effectively eliminate residual noise and further improve the output segmentation SNR.Finally,had written a file transfer program for data transmission using TCP/IP protocol.Based on the program,a speech enhancement experimental platform is designed,which use computer to collect data and DSP to complete signal processing,and the algorithm is transplanted to the DSP.The test results show that the improved method of this paper also realizes the function of speech enhancement on the experimental platform,and improves the segmentation SNR.