A template-based algorithm by geometric means for the automatic and efficient recognition of music chords

In this work, we introduce a template-based computational method to recognize chords through an audio recording of a musical instrument. The algorithm is based on a temporal frequency analysis using Gabor's filter banks. These filters are centered over adjusted frequencies of musical notes in different octaves and the adjustment is accomplished in terms of the detunings on the recording. Using the results in the filtering stage, a geometric mean of each chord is calculated. It is important to mention that these statistics are calculated from the combination of notes that form each chord and are automatically grouped as templates. The presence of chords is determined from these metrics. Several experiments are carried out for major, minor, augmented, diminished and suspended chords played on acoustic guitar, classic guitar, electric guitar, piano and ukulele. A comparative study against machine-learning classifiers is presented. The results show a superior performance of the present approach. In addition, the proposed method presents the advantage that it does not require a training stage, in contrast with the methods based on machine-learning algorithms. This reduces significatively the storage and time requiered for processing.