Research On Single Channel Speech Enhancement Applying Kalman Filter

Nowadays,everyone lives by connecting to the phone.It is extremely important to be able to keep the speech communication clear and smooth,wherever we are.This means that in most situations the noise environment where we are does not meet the most appropriate requirements for the conversation.In a system associated with voice communication,such as a telecommunications system or speech processing,the presence of background noise in a speech signal is undesirable.Background noise can make it difficult for the user to listen or reduce the performance of speech processing systems.Therefore,aiming at improving the quality of the speech signal,noise reduction is an important problem.In this thesis,we propose a method of single-channel noise reduction for speech enhancement.This method is based on the principle of spectral subtraction methods with the addition of a scalar Kalman filter to remove residual noise.The change in the speech spectrum is simulated as a Gaussian random process,and the residual noise is simulated in the form of Gaussian white noise for the application of the scalar Kalman filter.The Kalman filter used in this method is designed to represent the characteristics of the speech and noise signal.Our obtained experiment results with the online NOIZEUS speech corpus show that the proposed method has consistently improved the quality of the noisy speech.Besides,the experiment results also show that the PESQ and SNRseg improvement of the proposed method is improved compared with the other two basic implementations of spectral subtraction.