Chord Recognition using FFT Based Segment Averaging and Subsampling Feature Extraction

Abstract

This paper proposes a feature extraction subsystem for a chord recognition system, which gives a fewer number of feature extraction coefficients than the previous ones. The method of the proposed feature extraction is FFT (Fast Fourier Transform) based segment averaging and subsampling. Guitar chords were used in developing the proposed feature extraction. In general, the method of the proposed feature extraction is as follows. Firstly, the input signal is transformed using FFT. Secondly, the left half portion of the transformed signal is processed in succession using SHPS (Simplified Harmonic Product Spectrum), logarithmic scaling, segment averaging, and subsampling. The output of subsampling is the result of the proposed feature extraction. Based on the test results, the proposed feature extraction was quite efficient for use in a chord recognition system. For the recognition rate category above 98%, the chord recognition system only required a number of seven feature extraction coefficients. In addition, for the recognition rate category above 90%, the chord recognition system only required a number of six feature extraction coefficients.