The performance of DST-Wavelet feature extraction for guitar chord recognition

Abstract

Small systems can be designed to be more energy-efficient compared to larger systems. On small systems, the need for data processing with small data sizes becomes a necessity. In the context of small systems for guitar chord recognition, there are indications that further efforts can be made to reduce the size of feature extraction data. This paper introduces DST (Discrete Sine Transform)-Wavelet feature extraction to achieve this reduction. Basically, this work evaluated the frame blocking length, the number of DST cutting factors, and the type of wavelet filters (Daubechies and biorthogonal families) to obtain the optimal number of feature extraction data. Based on the evaluation, the optimal result obtained was a number of four feature extraction data. This optimal result was obtained by using a frame blocking length of 512 points, a DST cutting factor of 0.5, and a biorthogonal 3.3 wavelet filter. Testing with 140 test chords using these four feature extraction data could give an accuracy of up to 92.86%.