Speech Enhancement Based On Computational Auditory Scene Analysis

Speech enhancement algorithm can weaken the noise and increase the clarity of the target sound in the mixed sound, so the speech enhancement is paid attention to by scholars. At present, many methods can be used for speech enhancement, such as spectral subtraction, subspace algorithm and statistical model method and so on, The speech enhancement algorithm based on auditory scene analysis has important significance compared with other algorithms in the past, because it does not need to consider the voiceless and voiced, No matter what kind of noise can be filtered out. This thesis is based on the original speech enhancement algorithm based on computational auditory scene analysis, computational auditory scene analysis’ s abbreviation is CASA, in this algorithm,the wiener filter is introduced, and the quality of speech enhancement is analyzed.Firstly, the previous computational auditory scene analysis based on speech enhancement algorithm was studied, and effect of speech enhancement are analyzed, using the original algorithm enhanced voice although from subjective hearing could hear noise has a certain weakening, but the sound was harsh and the output signal to noise ratio of the system was not very high.Secondly, in order to enhances speech to noise ratio and make the sound more acceptable, introduces the improved wiener filter, through the combination of the auditory periphery model and wiener filter found using the improved algorithm can indeed improve the speech quality.Due to the method of combine the auditory periphery model and the Wiener filter has some limitations, we need a spectrum of known noise. Therefore, an algorithm is introduced to obtain the wiener filter system function by the noise spectrum estimation algorithm. The algorithm is used in the computation of the end of the improved algorithm based on computational auditory scene analysis, the algorithm can filter out the different types of noise. The algorithm is more concise, and it has a certain practicality.Finally, let CASA based speech enhancement algorithm, the auditory peripheral model combined with the wiener filter and improved algorithm based on noise spectrum estimation of CASA speech enhancement algorithm for the final three methods forcomparison, respectively from the subjective auditory, signal-to-noise ratio and log spectral distortion measure and waveform chart of the four aspects are compared, the results show the improved algorithm is superior to the original method, so that this improved method can remove the noise effectively.