cpd73 wrote: > Is that 1490 with just Musly? If so, that's slow - I used to get > ~20k/hour with just Musly, and ~1100/hour with Musly+Essentia. Also, why > recompile Musly? And if you did please use my fork - simply becasue I > merged in some pull request from the Musly repo. > > > > I've never actually got very far with Tensorflow. Looked into it, but > the models (AFAIR) are hundreds of megs. Just seemed overkill. Plus > Tensorflow, AFAIK, is slower - more accurate, but slower. > > > > First you will need to download the models you want to use. I'm not sure > if Essentia+Tensorflow produces the same attribute types as > Essentia+Gaia/SVM - probably does, but I can't be sure. > > > > MusicSimilarity just calls the music extractor, supplying a profile to > tell Essentia what models to use. Essentia then outputs a response in > JSON format (which MusicSimilarity optionally saves to disk). This JSON > is then parsed and the attributes extracted. yes, it's 1490/hour musli + essentia. I compiled your musly repo. As some people here have experienced, I didn't have all the dependencies installed when I ran it, so I didn't really investigate much and I just recompiled it! Have been doing this since 1995 with slackware, and do this for a living when I am bored :D Thanks for the explanations, I will try to get it running with tensorflow and post here if I am successfull, just for the fun of learning. Deep learning can do some incredible things, but it's a lot of timely trial and errors.
------------------------------------------------------------------------ ldarieut's Profile: http://forums.slimdevices.com/member.php?userid=21224 View this thread: http://forums.slimdevices.com/showthread.php?t=115609 _______________________________________________ plugins mailing list plugins@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/plugins
