Instrument Playing Technique Recognition: A Greek Music Use Case

Instrument playing technique recognition is a growing research field of Music Information Retrieval (MIR). Regarding stringed instruments, an instrument playing technique can be defined as any particular motive of the instrument players’ fingers applied on either the strings of the neck or of the body of the instrument. In this work, we examine the automated recognition of instrument playing techniques in solo recordings of the Greek stringed instrument bouzouki. To this end, a dataset comprising 500 audio segments belonging to 5 different playing techniques is generated. Hand-crafted audio features are extracted from each audio segment on a short-term basis, and then segment-level feature statistics are calculated as the final feature representation of each audio segment. Extensive experimentation for different types of classifiers and feature extraction optimization has been carried out to select the best model. Finally, the models are evaluated using an independent testing dataset, comprising 5 popular songs (whole recordings, not segments). To evaluate using these uninterrupted audio recordings, the non-existence of a playing technique (i.e. the “standard” way of playing a musical part, which has not been included in the initial classification task of the 5 playing techniques) is handled through a class-specific probabilistic thresholding of each class. An important contribution of this paper is that both the proposed classification pipeline and the dataset are openly provided for experimentation.