Multiple Audio Signal Separation And Identification Technology Research

In the real environment,the audio data signal will be influenced inevitably by environmental noise and other sound source signal,it bring many challenges in voice signal processing,in the meantime,finding and identifying the interested information in a large number of audio information is also a difficult problem.At the same time,it is one of the research hot spots in voice signal processing field in recent years.Therefore,the establishment of the audio signal separation and identification system has its extremely important theoretical research and application value in industry application,national defense military etc.Based on existing auditory physiology research,an auditory model is established,It mainly includes the outer-middle ear model that Can promote the audio signal in high frequency characteristics,and the basement membrane model which can be dealt with to multi-channel audio signal filtering,and Inner hair cells-auditory nerve model that can obtain characteristic information.In this paper,our characteristics include: the binaural time difference,binaural level difference,the auditory nerve distribution probability,auto correlation pattern and other parameters.By means of binaural time difference and binaural level difference and so on,to realize the separation of the audio signal.It elaborates systematic that the multiple frequency sound signal data by means of the Gamma tone filter to realize the separation of multiple frequency channels,calculating the sound source location azimuth in each frequency channel.The frequency channel could be divided into several big ownership category based on the differences of azimuth angle.According to auto correlation pattern in each frequency channel characteristic parameters,using the amplitude spectrum characteristics of the iterative algorithm to realize the signal reconstruction,and on the basis of the Half-wave rectifier inverse transformation technology to recover the lost negative signal part.In order to recovery achieve the frequency channel sound signal waveform.Multiple frequency channels of waveform figure which belongs to the same category to realize Lap joint together,and to obtain full waveform diagram of the audio signal.The waveform figure recovery process which belongs to other category is the same.Finally achieve the purpose of separation of audio signal.Then identifying the separated each audio signal.In this paper,the characteristics of twotypical are obtained in the process of extracting signal characteristic parameters,MFCC and sparse feature,then using support vector machine(SVM)to train the two features.At the same time,using the technology of feature weighting after the extraction of signal sparse characteristics,so that this feature can express the target signal fully,improving the recognition accuracy effectively.And eventually realize that by the SVM classifier to identify audio signal.After that,according to the whole system model algorithm,Using Matlab GUI simulation software to achieve the establishment of audio signal separation and the recognition systemFinally,it's the summary and outlook.As a whole in this paper,describing contents of separation and recognition system,and its advantages and disadvantages.As well as I have work done in the completion of the document,finally describes a task which is needed to be completed in the future.