Code used in ICASSP 2017 publication "Towards the characterization of singing styles in world music" by Maria Panteli, Rachel Bittner, Juan Pablo Bello, Simon Dixon
This project models attributes of the singing voice and explores singing style similarity in a set of world music recordings.
The code
- extracts features from melodic pitch contours
- classifies the contours as vocal or non-vocal
- learns a dictionary of singing elements from the vocal contours
- and groups together recordings with similar vocal content.
This project relies on motif: Melodic Object TranscrIption Framework and especially the "bitteli" feature extractor.