#metabrainz

Aerozol: can you share the yim 2024 logo to be used in the email? i tried exporting from figma but the quality is not good and there is a partial background.

lucifer: sending PM with the Gdrive link

thanks!

For posterity, I don't have the right images

Aerozol: i am looking for this image with a transparent background or with the color used in the YIM email mockups.

actually, looking at how the image and the text rendered in the test email. i think having the text in the image is a better idea.

monkey, mayhem, ansh, Aerozol : let me know if you notice any issues with the test YIM emails i just sent.

the header image and #YOUR YEAR IN MUSIC needs to be fixed but everything else renders fine for me.

yep, looks pretty good, lucifer

tuxpaint (IRC): sure, I can do that today.

what day are we planning to do the YIM run in prod?

Jan 1st??

cool, sounds good.

lucifer: the mbid-mapping container on gaga is now running the branch add-sentry-cron-monitoring . I've setup the sentry cron jobs and will start monitoring that things run appropriately.

great!

monkey: the & issue in yim images is fixed now.

mayhem: There once again seems to be a change to Apple's agreements that needs to be acceplted. Code signing is currently failing with a corresponding error message.

logged in, agreed.

clearly I read the thing carefully. :)

thanks.

of course, as we all do

lucifer: did anything change with our sentry setup? cron events are no longer being recorded.

the change would've happened on Dec 15th, which is when the test cron job stopped checking in.

monkey: update cover art mosaics made from all 2024 releases now updated on rex/rudi.

mayhem: not that i know of, checking.

I think sentry was rebooted near that time, no?

[listenbrainz-android] 1407jasjeet merged pull request #509 (03main…Fix-MissingAlbumArt-harsh_bhadauria): Fixed empty cover art showing blank space on brainzplayer & overview screens https://github.com/metabrainz/listenbrainz-andr...

mayhem: manually sending the checkin from my local PR works, manually running the cron command doesn't so likely an issue with decorator code or sentry initialization. checking that.

s/PR/PC/

unsure about that. the existing code that didn't use the decorator stopped working on Dec 15th.

well, maybe its that too, lol.

mayhem: i updated to latest sentry and it works auto-magically.

ahhh, did you update the container? if not, how do I update sentry?

requirements.txt in mbid_mapping directory. updated sentry there and built and deployed a new container

we'll know in an hour if it is working fine.

indeed, thanks!

looks like param found a place that serves momo pizza: https://x.com/iliekcomputers/status/18711950245...

amazing. want!

mayhem: the check-in happened fine.

sweet, thanks!

Hey all! Just joining the community in light of Spotify destroying all indie devs hopes and dreams with their recent API changes. Looking to get involved where do I begin?

hi igold94 !

what do you want to get involved with? there are lot of aspects to our projects. :)

Ya absolutely! Well I guess in part I don't want to be swimming against the current so to say. Does *Brainz have like a "roadmap" of sorts of where its intending to allocate its resources? Sure I have what I'd like from the platform but as a new contributor I need ramp up time before I could work any of those anyway

each project has it own roadmap, but the one for the ListenBrainz project is here: https://docs.google.com/document/d/1SG6gcYZnZ_V...

right now we're focusing on stability and new user onboarding -- trying to make all those better.

but in the new year I want to continue improving our similarity datasets and then improve LB radio more.

Listenbrainz and [unfortunately] AcousticBrainz were the projects I was the most interested in just for their utility. I do worry about the quality of recommendations from ListenBrainz that are purely CF-driven. During my time at Amazon Music I saw the pitfalls that can get you into and it's part of what makes Spotify's recs stand out so much (ie: that they're not solely CF driven)

Is there anything going on behind the scenes to come up with a more production-ready AcousticBrainz? If not happy to hop on the ListenBrainz band wagon

igold94[m]: To make a proper AcousticBrainz we would need a much larger team -- we teamed up with a uni who promised the algorithms were rock-solid. (hint: they were not). And they promosed to work on the project. (hint: they did not).

so AB is dead. CF is what we mostly have for recommendations at this point in time, which isn't great, but it isn't terrible either.

to bring more data into that recommendation process would indeed be great.

and we're using popularity and similarity data (and whatever else we can find) to build LB Radio: https://listenbrainz.org/explore/lb-radio/

including building a versions that works for local collections.

But is there a world in which if I did some research into acoustic features on the side that there would be a place for them in the *Brainz ecosystem? That's kind of where my passion lays

well!

yes, in particular we need a dataset of reliable BPM data.

Also general question about *Brainz - how do y'all get music from which to gather data? Like if I Wanted to do acoustic research or tagging or whatever where do I get the audio files? Does everyone provide their own?

AB is not reliable for that.

Well to be fair neither is spotify

they're actually quite terrible

igold94[m]: crowdsourcing. we could put these features into Picard for analysis.

very cool... so if I did some sploring into reliable BPM tools would that be of use potentially?

very much so.

but, I think that BPM is one of those tricky things - -there probably isn't one algorithm that is going to do it well for everything.

I was wondering if we should collect all the open implementations of BPM floating about and then run them all on data setset. then let vote for the correct one at the end.

Well ya I was gonna ask if there are mechanisms out there to get "voting" on correctness? Because I had a lot of issues with Spotify's BPM data it would very often be off by a factor of 2

I also had problems with their audio analysis dataset with beat/rhythm detection

yes, that is the nature of BPM detection. picking peaks in the frequency domains is rather tricky.

Ya now there are more human-like ways to infer BPM that require a better understanding of phrasing and higher level musical constructs which is also where my research interest very much lays

interesting.

Spotify is actually not bad at phrase and section detection which I'm sure is part of what makes their BPM detection better than average perhaps

well, you'd be a hero if you come up with a good way of detecting BPM

Great. Thanks for the pointers looking forward to getting into it

np. I'll tell ya what -- if you get an alg working, I'll work to get that run on millions of files. deal?

hahah deal

Is there a way to get audio files for analysis though or do I just have to obtain my research files on my own?

do you have your own music collection to start with?

that might be the easiest way of doing it for starters.

No but I can make one easy enough 😃

Also is Piccard the only way you guys do BPM tagging right now just so I can see where things are currently at?

great. for more extended testing, we can ask community members to run the code.

Ah v cool. very web3 lol

I dont think that picard does bpm tagging right now.

for lack of reliable data.

Okay and a few more questions here:... (full message at <https://matrix.chatbrainz.org/_matrix/media/v3/...>)

Just want to make sure I don't build away from a desirable direction

1. Ambient and classical. personally I think we should have "has beats" or "does not have beats" detection before we attempt to detect beats.

2. having the user select a "most representative section" and then running the alg on it. nothing fancier than that.

3. Where the data lives and how to deal with it: Not a problem, we'll handle it. If there is a good data, we'll find a way.

ansh: Hello! I think there is one thing that can be improved for the genre graph. Is it possible to remove the top-level user name group? Without it the colors come out a lot better (compare here top and bottom)

4. we dont have concrete structures or goals in place right now.

(sorry for the crosstalk)

mayhem[m]: If ambient and classical were the main problem areas (and I agree they should not have BPM data) then are there really systemic problems lol?