Automatic Labelling of Tabla Signals

Abstract

Most of the recent developments in the field of music indexing and music information retrieval are focused on western music. In this paper, we present an automatic music transcription system dedicated to Tabla - a North Indian percussion instrument. Our approach is based on three main steps: firstly, the audio signal is segmented in adjacent segments where each segment represents a single stroke. Secondly, rhythmic information such as relative durations are calculated using beat detection techniques. Finally, the transcription (recognition of the strokes) is performed by means of a statistical model based on Hidden Markov Model (HMM). The structure of this model is designed in order to represent the time dependencies between successives strokes and to take into account the specificities of the tabla score notation (transcription symbols may be context dependent). Realtime transcription of Tabla soli (or performances) with an error rate of 6.5% is made possible with this transcriber. The transcription system, along with some additional features such as sound synthesis or phrase correction, are integrated in a user-friendly environment called Tablascope.