Regarding crowdsourcing to improve metadata, the obvious success story is MusicBrainz. Although not the primary source I understand that Roon does use MusicBrainz so I guess if Roon users participate in improving MusicBrainz metadata then over time would get some benefit.
No I was saying it is better it is better to use SongKong before importing into Roon then to not. Having said that although the SongKong and Jaikoz matching algorithms are very similar currently the SongKong one is slightly ahead of the Jaikoz one, so you will likely get better results with SongKong then Jaikoz.
Actually, it’s working with Musicbrainz (and Discogs) data which has–in part–convinced us that even long-running “successful” crowdsourcing projects seem to always fall short.
Musicbrainz has a good schema, and decent set of curation standards (with exceptions–like the tendency to encourage placing composer names in the track artist field…a major lapse, IMO).
The size of the data set is rather small. Coverage of the long tail isn’t good at all (isn’t this what crowdsourcing should excel at?). Richness–in terms of populated track credits, reviews, bios, artwork–is a long way away from Rovi. Accuracy is about on par with Rovi…with biases towards differet kinds of errors in each.
The biggest problem IMO with MusicBrainz is that there is no way to automate adding of new releases, so new releases don’t make it in as quickly as they could. Also it is CD-centric so not so good for Vinyl, mainly because adding accurate track lengths is more difficult for Vinyl than a CD release. Discogs covers alot of data that MusicBrainz does not but its schema is not nearly so good. The absense of things like bios and reviews does not seem such an issue, e.g you only need one bio per artist which can be derived from other sources, the key thing is coverage of releases which as you say is not as large as Rovi. I am surprised you say long tail isn’t good, that has not been my experience. (Totally agree about composer names in track artist I mention that here About album identification problems)
If money was spent by commercial companies on actually adding data to MusicBrainz then I think some of the gaps could be plugged. There have been attempts by MusicBrainz to deal with Record companies directly but this has failed as they are generally very tied into Gracenote.
But you only have to look at TIDAL to see what happens with metadata from “commercial companies” (studios). Off the top of my head, would the real Sophie B. Hawkins please stand up?
And this is actually one of the more manageable metadata errors (a simple duplicate).
Oh yes I know Record labels don’t understand metadata, but at least if the release is added into an open system such as MusicBrainz then it can be improved/fixed by experienced editors who know what they are doing.
… and when I said Companies I mean’t companies such as Roon/Spotify ectera rather than Labels, i.e the companies that do value and use metadata.
Thanks for clarifying @paultaylor. It’s been some time since I trialed SongKong. But, my recollection is that it didn’t do a very good job of recognizing SACD/DVDA/Bluray versions of disks. So, I found myself turning to Jaikoz and MusicBrainz Picard where I seemingly had more control over choosing album/track versions. Maybe it’s user error though.
Its is true I haven’t done much testing with DVDA/Bluray because aren’t these Audio/Visual discs rather than just Audio and hence outside of the remit of SongKong ?
But if you are just ripping audio from audio/visual DVD then the resultant audio does not match a commercial audio product and can therefore not be automatched.
Okay so SongKong does use Acoustid, and in the absense of anything else SongKong does use the artist/title/album metadata that can be submitted with Acoustid. But this data is notorously unreliable so we only use if have no existing data, and if the acoustid tracks are not linked to a MusicBrainz release then we can only match Song Only rather than full Album match.
I’ve never understood why musicbrainz doesn’t have a bulk metadata upload facility. They could take bulk uploads, apply their rulesets and some machine learning to normalise and validate the submitted data, taking the good stuff and discarding the crud.
Wherever possible I’ve tagged with AMG metadata and there’s just no way I’m going to manually recapture metadata for musicbrainz’ benefit.
Totally agree, one of the oldest existing issues on MusicBrainz is https://tickets.metabrainz.org/browse/MBS-211 - their distrust of automatic editing verges on luddism, and is somewhat hypocritical since most MusicBrainz super editors use a series of scripts to semi-automate things. There was promising project to have a staging area where releases could be loaded and then imported into MusicBrainz but the developer lost interest and then it was canned when he left.
I have created http://www.albunack.net/ that makes it easy to find releases for an artist in Discogs that are missing in MusicBrainz and then you can import them one release at a time with a couple of clicks, but still somewhat laborious.
Yeah, entirely possible that I am doing something wrong. Maybe I’m doing something wrong in labeling or tagging my files before running them through SongKong. Or maybe I was doing something wrong with SongKong Are there some step by step instructions somewhere? Thanks!