Dear madmom team,
I would like to ask about the functionality of the SuperFlux function. I tried to use SuperFlux (madmom.features.onsets.superflux) using madmom but I have never achieved appropriate results. I also tried the standalone version (
https://github.com/CPJKU/madmom/blob/master/bin/SuperFlux) which worked fine. Haven't tried the NN superflux standalone yet.
I would like to ask you what is the proper calling of superflux() function via madmom.
I used:
_____
proc = mdm.OnsetPeakPickingProcessor (fps = 200)
act = mdm.SpectralOnsetProcessor (onset_method = 'superflux', fps = 200, filterbank = LogarithmicFilterbank, num_bands = 24, log = np.log10) (‘desired_wav_file.wav’)
onset_times = proc(act)
_____
For the superflux method, this function works relatively poorly (around 0.72 F-score for "Böck dataset", Boeck_ISMIR_2012), although according to the mentioned study it should reach higher values (moreover, this is the version using NN). The implementation in the librosa module also achieved better results on the selected dataset. I'd like to know if I'm making a mistake somewhere in the function call.
Thank you.
Best regards
Matej Istvanek