mherger wrote: > > First of all, thanks for releasing this plugin. I've been developing > a > > plugin for integrating EchoNest myself, but with a rather different > > approach (focusing on catalog radio playlists, and using taste > profiles > > as seeds), so it will be interesting to compare the results. > > Heh... where's the code? :-) > Well, it's not ready for the public, since it's pre-alpha quality. Additionally, I took a brute-force approach for the catalog radio playlists: - local music collection is not uploaded at all - playlist is created with a user-defined seed catalog - TEN's song suggestions are checked against local collection. If available, play. If not, tell TEN to ban artist and/or song and get a new suggestion. TEN seems to "learn" rather quickly, so the number of unavailable suggestions tends to decrease after a short time.
This works suprisingly well for a single user if you create a seed catalog which is representative of the local collection (~5 releases are enough), but will fail for a shared API key due to the amount of API calls it creates. So it probably makes no sense to release it at all, unless it's possible to get a non-commercial API key with a much higher rate limit. mherger wrote: > > > > - On both of the exports I ran (one with 0.94, the other 0.96), I got > > this error at the end: > > That's from the server.log, not the scanner.log, isn't it? It's some > postprocessing which is not part of the actual scan (though it is run > after the scan). > > > Since my collection is > 60000 tracks, I'm not sure if this was > finished > > sucessfully - maybe the same issue reported by slartibartfast. > > Check scanner.log for scanning issues. > Yes, this was from the server log. No errors in the scanner log, so I guess it should be fine... mherger wrote: > > > > it has gotten a lot better in the past months). One way to improve > this > > would be to let users optionally use their own API key (which is free > to > > get providing only minimal information to EchoNest), then the > > This has been suggested before. And it is explicitely forbidden by TEN. > I wasn't aware of that. Unfortunate, but kind of understandable from their position... mherger wrote: > > > > - For my collection, the recognition rate is about 30%, even though > it's > > quite well tagged (it contains mainly non-mainstream music, though). > One > > thing I noticed when experimenting with my plugin was that the > > recognition rate of songs really suffers from spelling differences > for > > artists (is it A Silver Mount Zion/A Silver Mt Zion/A Silver Mt. > > Zion/Thee Silver Mt. Zion Memorial Orchestra ? what about > > PJ.Harvey/P.J.Harvey/P.J. Harvey?). In order to improve this, one > could > > add an option to rename artists before the information is sent to EN. > > Rename based on what rule? > User-defined rules (e.g. "map PJ.Harvey to P.J.Harvey"). With my approach, you can easily see what song/artists TEN suggests (eg. Bob Mould Band instead of Bob Mould). With the SmartMix plugin, this won't be case so it would be a more manual process of discovery (as you noted in you post about using the blender for it) Roland ------------------------------------------------------------------------ Roland0's Profile: http://forums.slimdevices.com/member.php?userid=56808 View this thread: http://forums.slimdevices.com/showthread.php?t=97668 _______________________________________________ plugins mailing list plugins@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/plugins
