Acoustic Features for Determining Goodness of Tabla Strokes

The tabla is an essential component of the Hindustani classical music ensemble and therefore a popular choice with musical instrument learners. Early lessons typically target the mastering of individual strokes from the inventory of bols (spoken syllables corresponding to the distinct strokes) via training in the required articulatory gestures on the right and left drums. Exploiting the close links between the articulation, acoustics and perception of tabla strokes, this paper presents a study of the different timbral qualities that correspond to the correct articulation and to identified common misarticulations of the different bols. We present a dataset created out of correctly articulated and distinct categories of misarticulated strokes, all perceptually verified by an expert. We obtain a system that automatically labels a recording as a good or bad sound, and additionally identifies the precise nature of the misarticulation with a view to providing corrective feedback to the player. We find that acoustic features that are sensitive to the relatively small deviations from the good sound due to poorly articulated strokes are not necessarily the features that have proved successful in the recognition of strokes corresponding to distinct tabla bols as required for music transcription.