Automatic note transcription system for Hindustani classical music.

In Hindustani classical music, notes and their different variations play an important role to arouse the aesthetic qualities of a rãga. Therefore, detection of notes is very much important to find out the different characteristics of a rãga, but the task is very much challenging due to presence of improvisations or ornamentations. In this work, melody contour is extracted from the music file using salience-based predominant melody extraction method. Initially, the notes were determined by optimizing the tolerance band and duration of notes. The output of initial note transcription system consists of notes, its duration, and their boundaries (onset and offset instants). For improving the accuracy of initial note transcription, the melody contour is divided into melodic segments, and categorized into four broad categories based on duration and transition characteristics of the initial transcribed notes. Different features and classification models have been explored for classifying the melodic segments into desired and undesired categories. Further, we further proposed two metrics to measure the performance of the proposed transcription system. [ABSTRACT FROM AUTHOR]