Stringed musical instrument recognition using fractional fourier transform and linear discriminant analysis

This paper describes recognition of monophonic isolated sounds of stringed musical instruments using fractional fourier transform (FRFT) based MFCC features and Linear discriminant analysis (LDA). Performance of the system has been compared using conventional features like MFCC, Timbrel, Wavelet and Spectral features with proposed features based on FRFT and LDA. In proposed features FRFT has been substituted in place of DFT in Mel frequency cepstral coefficient (MFCC). FRFT, gives an additional degree of freedom of rotation of signal in time and frequency plane. Further, LDA implemented on these features enhances discriminant capability of these features. Feed forward neural network with back propagation algorithm was utilized and result were evaluated in terms of recognition accuracy. Eight stringed musical instruments with entire pitch range have been used to test the performance of the system. An accuracy of 94.37% gas been reported for eight stringed instrument recognition using FRFT based features and LDA against 75% for MFCC features.