Multiple Classification Recognition Based Instruments

With the popularization of information technology and the rapid development of Internet, it can be very convenient access to a lot of music, video and other multimedia data, with an urgent requirements of content-based multimedia information retrieval. Music signal must contain one or more characteristics of the acoustic, musical recognition is an important part of audio information retrieval.In this paper, the research goal is:Preliminary build a effective instrument classification recognition system; Master the instrument classification recognition process; Achieve the recognition of five kinds of Musical Instruments including piano, violin, flute, guzheng and pipa; Found a way to meet the requirement to identify the best classifier, research the selection of characteristic parameters, the combination of the various parameters influence on identification accuracy, and using semi-supervised learning method into musical instrument recognition. Based on the Weka experiment platform, using a variety of classifier, explore the effective way to improve instrument recognition rate.The main work in this paper includes:1. Collecting a variety of audio data via the Internet, including the establishment of piano, violin etc.,5kinds of small-scale music database.2. Researching and realization of musical audio feature extraction algorithm. Comparative and analysis LPCC, MFCC, DeltaMFCC feature parameters and parameter extraction through using MATLAB programming.3. Researching several kinds of classic classification algorithms in the instrument identification. By contrast experiment to find excellent performance from naive Bayes classifier algorithm, radial basis function neural network model, J48and random forest. The experimental results from the Weka platform show that with the optimal recognition results based on random forest classifier, the recognition rate up to95.7%. 4. Applied semi-supervised learning method into the instrument recognition, and using the Weka platform to realize its application.