Improved predominant melody extraction with a Source-Filter model (SIMM)

[juanjobosch]

Hi,

It would be interesting to have source-filter models (SIMM) implemented in essentia. The main motivation is the recently reported improvements over MELODIA. Specifically, it is interesting to perform novel combinations of salience functions, which have shown important improvements over Harmonic Summation shown in the following publications:

http://mtg.upf.edu/node/3569
http://mtg.upf.edu/node/3534

Python implementations (by J.L Durrieu) can be found here:

https://github.com/juanjobosch/SourceFilterContoursMelody/blob/master/src/SourceFilterModelSF.py
https://github.com/juanjobosch/SourceFilterContoursMelody/blob/master/src/SIMM.py

An implementation in C for the VAMP plugin can be found here:
https://github.com/wslihgt/IMMF0salience

An implementation of the combination of SIMM with Harmonic Summation, and the use with pitch contour selection is found here:

https://github.com/juanjobosch/SourceFilterContoursMelody