Guitar audio signal classification by collapsed Pitch Class Profile

Abstract

Guitar audio signal classification has its main application in chord transcription and guitar tutoring systems. This paper proposes a method to classify chords from an electric guitar. The method performs Wavelet Decomposition in order to split the signal in approximation coefficients and details, and then, those approximation coefficients below a threshold are removed. This filtered signal is reconstructed to apply Constant-Q transform, resulting in five octave length signal spectrum. The aforementioned octaves are further merged to a single one, and compared to prune the data. Next, the frequency bins with highest magnitude remain. Finally, the signal is passed through a classification step to perform chord recognition. Our proposed method outperforms some state of the art methods, with a simpler approach.