The Desgining Of Timbre Analysis System Based On Audio Signal Processing

There are variety kinds of music in the modern world.Music art is getting a greater and greater development with the rising of people's living standard.We can get different types of music anytime and anywhere.However,our request about the quality or others of music are getting up,too.New technologies about ASP(Audio Signal Processing)and Music retrieving and others of music are running faster and faster.As well,many scientists proposed instrument classifying based on the combination between ASP and Pattern Recognition.This is very important and have a great influence on music retrieving and music classification.We got start with our work from music timbre analysis,and designed a music timbre analysis system that based on ASP and combined with Pattern Recognition.Our mainly works showed as follows:Firstly,we extracted the features of timbre.We had preprocessed music signal based on ASP technology at first,such as for the noise reduction,we designed a group of band pass filters that response for the feel of human's ear to audio signal.Then we analyzed the features of music signal in the time-frequency domain,from piano to flute,we used eight kinds of instruments,we got some timbre's features in the time-frequency domain finally.At the last,we worked at feature extraction about timbre,based on the MPEG-7 standards and Mel-frequency Cepstral Coefficient,and made coding with MATLAB to implementing.Secondly,we combined the timbre's feature and Pattern Recognition to implement timbre of instruments classification,and four classifiers are used,they are nearest distance classifier,SVM(support vector machine),decision tree and back-propagation artificial neural network.We also tested our system on the WEKA platform that usually used for data mining.Then we had analyzed every classifier's good and bad points,this is necessary for our Timbre Analyze System's algorithm picking.Finally,we designed a timbre analysis system use Java and MySQL,this system can analyze the frequency domain of music,and after train some data,it can precision you input music is played by which instrument.This system also has some ability of self-regulation.