Research Of Music Instruments Identification Based On Acoustic Features

In recent years, with the rapid development of technologies in digital musiccreation, collection and storage, many agencies have accumulated a lot of music audiodata. It has become an urgent demand that how to effectively organize and managethese audio resources to make the people release from the work of querying a largenumber of audio data. In the sound source identification, speech signal processing andrecognition has already become a traditional hot spots. With the rapid development ofinformation science and technology, content-based audio and music signal analysis isbecoming a new research focus. Based on the acoustic features to describe the audiosounds, it is possible to achieve content-based music search engine queries based oncontent-based music features. The instrument identification is one of the importantapplications. It is widely used and applied based on the content of music transcription,structured coding of audio, music recommendation and the query engine and so on.This paper describes the basic theory and implementation of instrumentidentification based on the acoustic features. This paper describes the basic theory andimplementation of instrument identification based on the acoustic features. Firstly itmakes an used in-depth study in the instrument features and elaborates the featuresextraction methods. Then the system uses the MFCC,ΔMFCC as the featurescoefficients, focuses on support vector machine (SVM) classification principle anduse it as classification algorithm, then analyzes influence of the different acousticfeatures to the correct rate of instrument identification.Finally, the paper uses PCA and base-largest interval improved SVM featuresselection algorithm to do the further experiments. And the results show that PCA isable to simplify the feature vectors and the improved feature selection algorithm canfurther improve the recognition accuracy of the instrument identification.