Study And Hardware Realization On Multi-Microphone Speech Enhancement Methods

In the process of collecting audio signals, such as in the audio communication system and video phone conference system etc., interference of environment is unavoidable, which always depresses the communication quality, even could make the audio signal processing system out of work. For many years, researchers are trying their best to study on various excellent speech enhancement methods, in order to reduce environmental noise and promote the communication quality in the speech enhancement process.Based on previous researchers' efforts and achievements, this thesis is mainly about following two aspects of study on speech enhancement methods, and a realization of designing on 4-microphone audio signals collecting systems with TMS320VC5509 DSP.Empirical Mode Decomposition (EMD) speech enhancement method by shrinkage function post-process is studied. EMD method separates waves or trends of different scales from the signal step by step, and then expresses the signal with instantaneous frequency and energy. An improved method which separates noise component with EMD speech enhancement method and then promotes its noise reduction performance with shrinkage function post-process is proposed. The simulation results showed the validity of the proposed method.Multi-Microphone Signal Subspace speech enhancement method based on average Kurtosis Values Voice Activity Detection(VAD) method in subband is studied. First of all, VAD results of the input noisy signals are determined according to the mean of Kurtosis Value in all the sub-band signals frame by frame, and then the voice frames are processed by Multi-Microphone Signal Subspace(MMSS) method, which can reduce the computation load of MMSS method. At the same time, the formula of short-time SNR based on Single-Microphone Signal Subspace method is modified to estimate Lagrange Multiplier accurately, and then to improve the performance of speech enhancement in the Multi-Microphone Signal Subspace method. The simulation results showed the validity of the proposed method.4-channel audio signal collecting system realized by adding a Codec module based on SEED-DSK5509 provides a hardware platform for theory research and application for Multi-Microphone speech enhancement methods.