The AB update sounds like a huge step, I'm always wondering if there's a way to mark recordings that are variations of the same theme, not sure if this will help, gonna play with it a little bit
2021-09-01 24433, 2021
alastairp
yyoung[m]: this won't work with variations, we don't have enough information in the acousticbrainz data to identify that
2021-09-01 24453, 2021
yyoung[m]
Oh, OK :)
2021-09-01 24407, 2021
alastairp
there's a subbranch of this area of research called "cover song identification", which mostly works with identifying different recordings of the same song (i.e. live/studio/cover)
2021-09-01 24445, 2021
alastairp
but variations is interesting... you'd probably have to identify the melody, and then on top of that decide what a "variation" actually means
2021-09-01 24432, 2021
yyoung[m]
For example, a slower or softer version, or an arrangement?
2021-09-01 24404, 2021
alastairp
cover song identification should help with slower or acoustic versions
2021-09-01 24421, 2021
alastairp
different arrangements - often as long as the general chord structure is the same it should also work
2021-09-01 24405, 2021
yyoung[m]
Yes, at long as we can identify melodies that would be near success
2021-09-01 24452, 2021
yyoung[m]
Why are there 2 "MFCCs"s?
2021-09-01 24451, 2021
alastairp
the one that ends with w is weighted
2021-09-01 24417, 2021
alastairp
so they're normalised to have values between 0 and 1
2021-09-01 24437, 2021
alastairp
ideally this means that things that are otherwise the same, but one is louder than the other will still appear as identical
2021-09-01 24459, 2021
alastairp
because if we didn't weight them, it would identify them as "different", because the numbers that it compares would be different
2021-09-01 24412, 2021
yyoung[m]
I mean, the last one in the list is also "MFCCs", so there're MFCCs, MFCCs(weighted), MFCCs
I am stupid. I check this query on artist without urls (:
2021-09-01 24438, 2021
yyoung[m]
I'm curious about how instrument similarity is measured š¤
2021-09-01 24424, 2021
alastairp
yyoung[m]: this name is a bit misleading, and something that we can improve
2021-09-01 24444, 2021
alastairp
we have a machine learning algorithm which tries to identify if a recording is "instrumental" (i.e. doesn't have human vocals)
2021-09-01 24423, 2021
yyoung[m]
Right, and that's not quite accurate according to my testing :)
2021-09-01 24434, 2021
alastairp
and the result is something like "81% instrumental, 19% vocal"
2021-09-01 24453, 2021
alastairp
and then the idea is that a recording which is identified as 80.5% instrumental, 19.5% vocal would be "similar"
2021-09-01 24421, 2021
alastairp
but yes, of course there is also the question of if the underlying model for instrumental/vocal is good
2021-09-01 24429, 2021
yyoung[m]
Ah that's misleading indeed :D
2021-09-01 24451, 2021
yyoung[m]
And I'm not sure if this metrics makes any sense
2021-09-01 24459, 2021
alastairp
in this case, perhaps it doesn't make too much sense to use such a complex "similarity" metric for this. in reality you probably just want to group all items with a "instrumental" % of > 70% into a single group
2021-09-01 24408, 2021
alastairp
however
2021-09-01 24430, 2021
alastairp
next step is to look and see if it makes sense to "combine" these metrics and use more than one at a time
2021-09-01 24455, 2021
alastairp
perhaps it makes sense to be able to search based on 1 similarity and then filter based on another?
2021-09-01 24433, 2021
yyoung[m]
It's already beyond my expectation though, seeing it shows 3 songs I've heard before
2021-09-01 24442, 2021
yyoung[m]
Oh nice, seems it can identify short versions
2021-09-01 24424, 2021
Etua joined the channel
2021-09-01 24404, 2021
Etua has quit
2021-09-01 24440, 2021
alastairp
yyoung[m]: if you have ideas about what to do with it, then let us know! we made the feature but don't know what to do with it
2021-09-01 24450, 2021
alastairp
now that the similarity is finished, I can work on dumps for MrClon!
Congratulations and thank you to all of this year's GSoC students for all your hard work ! Looking forward to seeing you around in our community akshaaatt[m] , akashgp09[m] , jasondk , yyoung[m] and ritiek !
2021-09-01 24414, 2021
yyoung[m]
monkey: Thanks. And thank yvanzo for the final evaluation, that's really nice.
2021-09-01 24455, 2021
BrainzGit
[bookbrainz-site] 14MonkeyDo merged pull request #618 (03masterā¦edition-physical-data): [BB-621]: Physical data section should be disabled if eBook is chosen https://github.com/bookbrainz/bookbrainz-site/pulā¦
2021-09-01 24429, 2021
akshaaatt[m]
Thank you monkey ! The whole team at meb has been really amazing!šÆš
2021-09-01 24421, 2021
akshaaatt[m]
lucifer thank you for the kind words as always. You did deserve to become god. š„
2021-09-01 24424, 2021
reosarevok
yyoung[m]: I think closing that ticket would make sense, yes
2021-09-01 24404, 2021
reosarevok
Either as a duplicate, or mark it as on beta if it's not a clear dupe of anything
2021-09-01 24442, 2021
reosarevok
lucifer: did it finish yet?
2021-09-01 24410, 2021
lucifer
reosarevok: artist, release done. recordings going on now.
2021-09-01 24416, 2021
reosarevok
bitmap, yvanzo: I'm in the countryside without a charger, so around for however long my battery lets me
2021-09-01 24426, 2021
reosarevok
If you want to discuss stuff :)
2021-09-01 24451, 2021
reosarevok
We could look at the remaining React edits maybe, to get that over with, if you have no other preference?
2021-09-01 24400, 2021
reosarevok
lucifer: oh, ok, so... tomorrow I guess? :D
2021-09-01 24415, 2021
lucifer
yup, my guess too
2021-09-01 24412, 2021
bitmap
reosarevok: I'm good to look at react edits in a few mins
2021-09-01 24434, 2021
yvanzo
moāāinā
2021-09-01 24450, 2021
yyoung[m]
reosarevok: I submitted a patch of 3 beta tickets
2021-09-01 24444, 2021
reosarevok
Oh
2021-09-01 24450, 2021
reosarevok
That seems more important then
2021-09-01 24454, 2021
reosarevok
I'll start there
2021-09-01 24427, 2021
yvanzo doubles monkeyās thanks and congratulations to all GSoC students!
2021-09-01 24410, 2021
yvanzo
lucifer, reosarevok: Feel free to use sir-test on aretha, I just needed it for yesterdayās release.
2021-09-01 24412, 2021
yvanzo
lucifer, reosarevok: Feel free to use sir-test on aretha, I just needed it for yesterdayās release.
MBS-11931: Beta: Link type option "Discogs" appears under the "get the music" heading
2021-09-01 24426, 2021
reosarevok
(I kinda expected it to show the otherdbs heading but actually this seems fine too, just making sure)
2021-09-01 24441, 2021
yyoung[m]
"Discogs" is under "otherdatabases", but that's not allowed
2021-09-01 24443, 2021
reosarevok
Sure :) Just making sure
2021-09-01 24414, 2021
reosarevok
Seems fine to hide that anyway, I expect the headers are rarely if ever useful as, well, headers when we have a small selection
2021-09-01 24426, 2021
reosarevok
These look good to me, if yvanzo likes the code changes (I mostly tested onsite)
2021-09-01 24440, 2021
reosarevok
yvanzo: did you see my ping on #mb earlier about the url guideline, btw?
2021-09-01 24446, 2021
Toasty joined the channel
2021-09-01 24457, 2021
Toasty has quit
2021-09-01 24410, 2021
reosarevok
My battery is starting to run out, but I don't think I have a lot more to say than what is already there :) So I'll just continue from my phone if we do some reviews involving my code
2021-09-01 24448, 2021
MrClon
I try wrote script to mark as "ended" bunch of broken bandcamp links i found, but it is harder then i thought. I'm afraid I might break some data in DB. Can i somehow use MusicBrainz Batch Edits for it? I can prepare list of broken links in any format
But i can't check 8000 edits to ensure there is no some edge cases i did not think about
2021-09-01 24441, 2021
reosarevok
Sure. You should only enter edits you're fairly sure about, though, and a bot is limited to 1000 edits per day specifically so that it's not impossible for users to check them :)
2021-09-01 24435, 2021
Toasty joined the channel
2021-09-01 24457, 2021
MrClon
I have problem not with Ā«sure aboutĀ» part, but with Ā«enter editsĀ»
2021-09-01 24439, 2021
yyoung[m]
bitmap: Could you review the PR again?
2021-09-01 24432, 2021
bitmap
sure
2021-09-01 24420, 2021
reosarevok
MrClon: does the bot code help in any way? That does have ways already written to set relationships as ended based on a SQL query, basing it on a static list shouldn't be super hard, I expect
2021-09-01 24447, 2021
reosarevok is off for the evening, more tomorrow once I get back to civilization aka my charger
2021-09-01 24402, 2021
yvanzo
Good evening reosarevok!
2021-09-01 24455, 2021
yvanzo
MrClon: Sure, you can create a list of broken links in an MBBE ticket.
2021-09-01 24416, 2021
yvanzo
Testing the PR again.
2021-09-01 24449, 2021
MrClon
reosarevok: will look at it
2021-09-01 24431, 2021
MrClon
Similarity don't work on AB. Hang on endless Ā«Loadingā¦Ā», in Firefox console `TypeError: t.metadata[0] is undefined` (similarity.b24730c55c2152da173f.js:2:134104) in Chromium `Uncaught TypeError: Cannot read property 'youtube_query' of undefined`
that youtube error looks unrelated. this seems to be the issue alastairp mentioned earlier. an automated job needs to be setup to keep the annoy similarity index up to date.
