Performance of DCT-DWT Feature Extraction in Recognizing Guitar Chords

This work deals with optimization problems in the field of audio signals with particular emphasis on recognizing guitar chords. As an extension to the previous works done in this field, the primary goal was to minimize the feature extraction length without loss of recognition performance. In this case, this work used two such well known methods in audio signal processing as the Discrete Cosine Transform (DCT) and the Discrete Wavelet Transform (DWT). This method is based on the iterative analysis of three main system attributes: the frame blocking length; the choice of wavelet functions; and the DWT coefficient selection offset. After careful tuning of these parameters, this work was able to optimize the feature extraction length to only three, which however represents a modest improvement on the previous works in this field. The system developed in this research obtained recognition accuracy of up to 91.43% when the optimized feature extraction length was employed. The significance of this study extends to the non-conventional applications of guitar chord recognition. Shortening the feature extraction length is expected to improve more usages in the different applications of the music technology.