Designing automatic note transcription system for Hindustani classical music

Hindustani music is heterophonic with lead voice accompanied by instruments. A trained Hindustani musician is capable of perceiving the notes based on the lead voice but a novice person is unable to decode the notes. This necessitates the development of an automated note transcription system. The automatic system detects and generates the notes present in the music file. In this work, the melody contour is extracted from the audio file using salience based method. The extracted melody values are normalized in cent scale. The notes are having a fixed melody value in cent scale. Each of the melody values from the extracted melody contour is compared with the melody value of a note. If the extracted melody value matches within a given range of tolerance of melody value of a note then this note is assigned. The successive similar notes for each melody value of the contour are merged together to form a single note preserving the start and end time. In the method, some notes are also detected during transition of melody contour from one note to another. These notes are termed as transitions notes which are not perceived and not desired. The durations of these notes are very less. These notes are eliminated by using duration thresholding. So, the tolerance of melody value and threshold duration of the note plays an important role in the accuracy of the transcription system. These parameters are optimized to maximize the accuracy of the system. The performance of the system is evaluated using two metrics. The results show the note transcription system performs satisfactorily.