Modelling of Gamakas for Karnatic Flute Music Synthesis

Abstract

In this work, we propose a spectral model to efficiently synthesise Karnatic bamboo flute music from the notes, duration, and raga information of a song. Karnatic flute music synthesis from basic notations is a challenging problem due to two major reasons. The first one is that the gamakas are generally omitted from the musical notations in the tradition of KM. Hence, for the automatic synthesis of KM, the gamakas associated with every note need to be predicted from the musical notations. The second reason is the continuously varying pitch contour of a note in the presence of gamakas. We propose a method to detect the presence and type of gamakas associated with each note in a data-driven manner, from the annotated symbolic music alone. In this regard, we propose features based on the notes of the song. These features are used as inputs to a Random Forest Classifier (RFC). From our experiments, the accuracy values obtained for predicting the presence and type of gamakas are 77% and 70%, respectively. These are significantly better than random classification accuracies. We also analyse the importance of neighbourhood of notes for the detection and classi cation of gamakas. It is observed that the best accuracy is obtained for gamaka presence detection when a both-sided neighbourhood of size three is considered; and the best accuracy for gamaka type prediction is obtained with a both-sided neighbourhood of size one. The analysis performed on the training data reveals that there is information contained in these neighbourhoods for distinguishing between gamaka and non-gamaka notes. For synthesising Kar _ n at.ic ute music, we model three di erent components of the ute sound, namely, pitch contour, harmonic weights, and time domain amplitude envelope. Cubic splines are used to parametrically represent these components. Subjective analysis of the results shows that the proposed method is better than the existing popular methods in terms of tonal quality as well as the propriety of rendering gamakas. Hypothesis test results show that the observed improvements over other methods are statistically signi cant at 95% con dence interval.