Improvement Of Audio Feature Extraction And Research On Visualization Method

With the development of technology and Internet,human life data shows explosive growth,and sound(voice,music etc.)as an amount of data in the life of human society,processing and analysis of massive sound has become a daunting challenge.Data visualization technology is a simple and effective way to summarize the data,it can improve the traditional sound analysis methods(such as spectrum)of low efficiency,high professional barriers problem.The main content of this paper is audio visualization,which includes the audio features extraction and the visual elements presentation,and this paper proposed a solution of audio visualization system,they are as follows:1.When extracting sound features,first of all,the sound should be converted into digital audio which can be processed by the computer,and then the noise reduction,framing windowing,extracting short-term and global features at last.Pitch as the basic physical properties of sound,reliable accuracy of pitch extraction will directly affect the subsequent visualization results,and the accuracy of the current study on pitch extraction also needs to be further improved,therefore,this paper proposes SHI method for pitch estimation,the method mainly includes SHI function transformation and period detection of two steps,SHI function transformation can transform the original waveform to a waveform with more obvious periodicity compared to ACF,AMDF etc.,and period detection is that calculating the weights of the hypothetical period,and take the period value with the largest weight as the final result.SHI method for pitch estimation has better experimental effect compared to the ACF and the other methods.Using SHI method can reduce the error rate of pitch extraction to a half compared to other methods,such as YIN.2.In order to improve the accuracy of subsequent tone and melody extraction,this paper also proposes the matrix graph filtering algorithm,the core idea of the method is converting a numerical sequence of 1D into a 2D matrix and matrix graph structure,and then through horizontal expansion of matrix,decision analysis of communication path and the relative relationship between the numerical in space to calculated a context number.The experimental results show that the method can not only compress the representation of the data effectively,but also improve the accuracy of tone estimation.Compared to some traditional filters,such as median filter,matrix graph filter can improve SNR of sequence by 50% nearly.3.Based on the extraction of features,mapping auditory perception to coincident graphic elements,then drawing and transferring these elements in a reasonable way,paper presents an audio visualization system B/S architecture scheme at the end.