<02UltimateRiff> @_flaky should be excited about that one~
2024-06-14 16615, 2024
beanbrain has quit
2024-06-14 16645, 2024
beanbrain joined the channel
2024-06-14 16645, 2024
beanbrain has quit
2024-06-14 16645, 2024
beanbrain joined the channel
2024-06-14 16631, 2024
tarulia joined the channel
2024-06-14 16622, 2024
tarulia has quit
2024-06-14 16640, 2024
beanbrain has quit
2024-06-14 16658, 2024
discordbrainz
<09duckfromdisc> whats good yall
2024-06-14 16641, 2024
discordbrainz
<09duckfromdisc> so i'm really passionate about what acousticbrainz used to be
2024-06-14 16607, 2024
discordbrainz
<09duckfromdisc> anyone know if there's something similar going on where it's a bunch of music reduced to vectors in a ???-dimensional space and linked to MBIDs?
2024-06-14 16601, 2024
discordbrainz
<09duckfromdisc> i get this reference
2024-06-14 16652, 2024
discordbrainz
<09duckfromdisc> also another question, are there any alternate AB servers around?
2024-06-14 16633, 2024
discordbrainz
<09duckfromdisc> acousticbrainz can be hosted through docker i've heard and since the main one was discontinued i was wondering if there was a continuation somewhere else that is accepting new data
2024-06-14 16621, 2024
discordbrainz
<02UltimateRiff> @duckfromdisc I think #development (bridged to #MetaBrainz) or the forums might be a better place to ask about AcousticBrainz
2024-06-14 16625, 2024
discordbrainz
<09duckfromdisc> đź‘Ť
2024-06-14 16649, 2024
discordbrainz
<03_flaky> @UltimateRiff hell yeah
2024-06-14 16655, 2024
discordbrainz
<04elomatreb> I personally wouldn't expect this to be a thing fwiw, the data generated by/through AB just wasn't that useful
2024-06-14 16610, 2024
discordbrainz
<09duckfromdisc> yeah
2024-06-14 16620, 2024
discordbrainz
<09duckfromdisc> was wondering if there was some other project that basically did the same things but with "better" models
<09duckfromdisc> oh wow that was a terrible example
2024-06-14 16604, 2024
discordbrainz
<09duckfromdisc> that just loaded on my end, it's entirely copies of the same song
2024-06-14 16614, 2024
discordbrainz
<04elomatreb> ah, right
2024-06-14 16620, 2024
discordbrainz
<03metadataismydrug> " Gammatone Frequency Cepstral Coefficients"?
2024-06-14 16638, 2024
discordbrainz
<04elomatreb> that sounds like something a guy on star trek would say tbh
2024-06-14 16646, 2024
discordbrainz
<10volumetrique> LOL
2024-06-14 16610, 2024
tarulia[m] joined the channel
2024-06-14 16610, 2024
tarulia[m]
and probably did
2024-06-14 16633, 2024
discordbrainz
<04elomatreb> Specifically for the recommendation side of things, LB aims to "take over" some of those goals (just based on listen cohort similarity instead of audio-level similarity)
2024-06-14 16600, 2024
discordbrainz
<10volumetrique> From the few recordings I've tried, AB was really flawed
<10volumetrique> I'm confident that just matching people's listen like elomatreb is a better way - isn't that how most streaming algos work and how YT also works... ?
2024-06-14 16652, 2024
discordbrainz
<09duckfromdisc> thats why i was hoping perhaps someone made a better version using MBIDs but with decent models
2024-06-14 16657, 2024
discordbrainz
<04elomatreb> the problem is that it only works as long as there are sufficiently many listens to establish a proper grouping
2024-06-14 16608, 2024
discordbrainz
<04elomatreb> i.e. it quickly starts failing for obscure music
2024-06-14 16614, 2024
discordbrainz
<09duckfromdisc> which becomes really difficult when you want to discover up and coming artists
2024-06-14 16625, 2024
discordbrainz
<04elomatreb> where just one active listener is enough to significantly skew the statistics
2024-06-14 16654, 2024
discordbrainz
<09duckfromdisc> what's elomatreb? as far as i know spotify uses the audio features and vector spaces
2024-06-14 16605, 2024
discordbrainz
<10volumetrique> I feel like model based predicting has never been able to succeed so far
2024-06-14 16612, 2024
discordbrainz
<04elomatreb> elomatreb is me
2024-06-14 16616, 2024
discordbrainz
<10volumetrique> Plex's sonic analysis isn't great either
2024-06-14 16622, 2024
discordbrainz
<09duckfromdisc> o
2024-06-14 16629, 2024
discordbrainz
<09duckfromdisc> i thought they meant you had an algo fo rit
2024-06-14 16654, 2024
discordbrainz
<03metadataismydrug> I'm pretty sure Plex's system is a lie and they just use metadata
<09duckfromdisc> a small dataset and i also feel like the models used are slightly outdated/arbitrary
2024-06-14 16638, 2024
discordbrainz
<09duckfromdisc> there are a lot more genres than just 8
2024-06-14 16606, 2024
discordbrainz
<09duckfromdisc> if you look at tunebat spotify models make more sense
2024-06-14 16629, 2024
discordbrainz
<09duckfromdisc> i guess we could use LB for that then
2024-06-14 16641, 2024
discordbrainz
<09duckfromdisc> i wanted to write a "recommendations engine" for maloja
2024-06-14 16654, 2024
discordbrainz
<04elomatreb> If it didn't take so much electricity and/or money for nvidia I would be interested in trying to apply some of the tech behind the recent "AI" hype to music analysis
2024-06-14 16600, 2024
discordbrainz
<09duckfromdisc> that's what im saying
2024-06-14 16627, 2024
discordbrainz
<09duckfromdisc> i was discussing recommendation engines with someone in the maloja server and i mentioned acousticbrainz and i said it's not a matter of if it's possible, it's a matter of how much time/power/data/internet it consumes
2024-06-14 16604, 2024
discordbrainz
<09duckfromdisc> it's 100% possible to apply AI to a bunch of songs and put them into a vector space. what's difficult is finding the processing power to do so
2024-06-14 16617, 2024
discordbrainz
<04elomatreb> although IMO the best recommendation engine in terms of actually new discoveries and interesting stuff are music journalists/blogs/etc
2024-06-14 16623, 2024
discordbrainz
<04elomatreb> just harder to scale
2024-06-14 16636, 2024
discordbrainz
<09duckfromdisc> i think i'm the best recommendation engine
2024-06-14 16645, 2024
discordbrainz
<09duckfromdisc> but i only accept two songs
2024-06-14 16649, 2024
discordbrainz
<09duckfromdisc> i guess the idea would be something like applying lb-radio (what is troi?) to the maloja api
