Monophonic Music Recognition Based On Extreme Learning Machine And Deep Belief Net

Musical signal processing is a significant component in the signal processing field.Musical signal is a special quasi-periodic signal, and it has more abundant tone-color,more complicated structure of frequency compared with audio signal, and the range of its spectrum is more extensive, and the rhythm feather of its time domain is more obvious.Here we study the musical recognition combined with musical theories and computer media technology,signal processing and the related knowledge of pattern recognition.We also use computer to recognize 88 monophonic music signals.In this paper,we firstly introduce the development of computer music,and describe the basic musical theory and signal feature.Secondly, we introduce a transforming tool called CQT which transforms musical signals from the time domain to the frequency domain.We also improve rapid CQT method and rapid CQT in frequency domain to reduce its complexity.In order to recognize the piano monosyllabic signal,we present some basic theories about single-hidden layer feedforword neural network, so-called extreme learning machine. Finally, we introduce RBM and DBN which has many RBM, as well as two kinds of classification. Which we used to recognize the piano monophonic music signals.This paper extracts features from 88 monophonic music signals with.wav format using CQT method,and utilizes these features and their labels as training samples of extreme learning and DBN. Next, we use extreme learning and DBN to learn, and use two kinds of method to recognize 88 monophonic music signals.Extreme learning machine has a better accuracy rate of 92.81%. This illustrates extreme learning machine is valid to the recognition of monophonic music signals.