Research And Application Of Speech Enhancement Algorithm Based On Improved Spectral Subtraction

In people's daily life, noise from surrounding environment could severely interfere audio signals, which negatively affect their further samplings and processings. In this way,my work is to decay noises from outside by the enhancement of the speech signals so that high quality of audio signals could be achieved.This paper mainly focus on the algorithm or the improvement of audio signals, and the system for the recognition of audio signals based on Android platform. In previous works about traditional spectral subtraction, we note that it is hard to achieve perfect signals and the extra introduction of musical noises will also bring us new problems. In this work, we try to use spectra subtraction together with MSCPE noise estimation method to acquire useful signals among several noise signals. The majority work are stated as follows:First of all, on the basis of noise producing model and characteristic, in order to get speech signals with different noises. We add several different noises, which come from NOISEX-92, in different circumstances of noise ratio. And using these signal as process objects of different algorithms mentioned before.Secondly, this paper has researched three methods of enhancing speech signals, and use all the methods to processing the signals obtained before. What is more, to support further study, it completed the comparison in signal enhancement performance among these signals.Thirdly, through analysing the characteristic of human ear, this paper choose spectral subtraction as the major research topic, the combination of spectral subtraction, MESCPE and ear masking effect have been employed as effective method for the enhancement of speech signals — — MSCPE spectral subtraction based on the ear masking effect.Furthermore, the comparison experiments between this new algorithm and traditional algorithm proved that this method can not only help researchers to obtain pure speech signals, but also decrease the music noise caused in traditional spectral subtraction. It further proved the superiority of this new algorithm in enhancing speech signals.This paper establish a speech recognition system on Matlab platform, based on dynamic time warping model, and perform several tests on this system. It has been proved that this algorithm mentioned in this paper can improve the distinguish rate in some extent,through compare the performance of different algorithms which used to enhancing speech signal.Finally, through all the works mentioned above, this paper design and achieve a signal recognition system based on Android terminal. All the test performed on this plat proved the effectiveness in decreasing noise ratio as well as the superiority in enhancing distinguish ratio.