Audio Architecture Technology Research

The audio frequency structurization is a process which cuts the audio frequency into independent and stable unit, and then acquires the related scene sorts of the units by data analysis. the frequency structurization is not only beneficial for the deep analysis and processing of audio, it also plays an important assistant role on the video analysis and searches based on contents. The paper, which focuses on the research on related technical points of audio frequency structurization that including the abstract, division and classification of audio, has make the achievements as follow:Firstly,For the abstract of audio characteristic, the paper introduces the MFCC characteristics based on UBM and the tamber characteristic of internal and external frame standard deviation. the experiment result proves the validity of the new character. the selection of original characteristics through orthogonal experimental design provide a identity reference base for distinguishing programs of different audio sorts.Secondly, the paper introduces a audio partition calculation methods base on the changing reliability tests. The new method adapts a fixed length gliding window testing structure which can reduce the cumulated mistakes. To calculate the reliability of each audio point inside the window, and then test the trip point by based on the trend of reliability, which in order to avoid the mischeck by thselect and threshold. The experiment proves the advantages of the partition based on the new calculation.Finally,the paper also introduces the audio classification calculation based on VQ-GMM. To classify roughly and then recognize accurately based on the structure character of audio. And the experiment result shows that the classification based on the new calculation method provides a more precise way than hierarchical method, the nearest feature line and VQ classification.