[HN Gopher] AcousticBrainz: Making a hard decision to end the pr...
___________________________________________________________________
AcousticBrainz: Making a hard decision to end the project (2022)
Author : petemir
Score : 149 points
Date : 2023-01-19 11:37 UTC (11 hours ago)
(HTM) web link (blog.metabrainz.org)
(TXT) w3m dump (blog.metabrainz.org)
| rom-antics wrote:
| > Unfortunately, the resolution of the data that we collect in
| AcousticBrainz is not enough to be used in this type of machine
| learning, and so we were unable to try these new techniques using
| the data that we had available in the database.
|
| Reading between the lines here -- AcousticBrainz can't store
| copies of any actual audio for copyright reasons. Instead they
| store various fingerprints and derived properties of the audio
| signal (example[1]). When the project started they didn't
| anticipate the right kind of input data required for these newest
| AI techniques, so to make progress they would have to throw away
| 7 years of data and start over their data acquisition from
| scratch. Instead they decided to shut it down.
|
| Just another way the RIAA is making the world a worse place to
| live.
|
| [1]
| https://acousticbrainz.org/api/v1/6df2d3b5-25e1-4d4c-a196-04...
| CharlesW wrote:
| According to TFA, even if they had copies of every song they'd
| ever analyzed they'd still need to re-start from scratch. The
| reason they're not is:
|
| "...we don't have the resources and developer availability to
| perform this kind of research ourselves, and so we rely on the
| assistance of other researchers and volunteers to help us
| integrate new tools into AcousticBrainz, which is a
| relationship that we haven't managed to build over the last few
| years."
|
| If the RIAA was to blame, Pandora's Music Genome Project and
| other music analysis/recommendation systems wouldn't exist.
| wpietri wrote:
| I think your main point is correct, but the Music Genome
| Project had experts humans listening to each and every song.
| It originally started out to build music-store recommendation
| kiosks (yes kids, music used to be sold in physical stores),
| and it took them 5 years to pivot to streaming audio. Given
| when they started, I expect that they just bought a lot of
| CDs to start. By the time they got into streaming, they of
| course had licenses. So Pandora didn't defy RIAA; they spent
| money and built business relationships, two things
| unavailable here.
| Cthulhu_ wrote:
| I know Spotify does a lot of analysis and clever things with
| the music on their platform, but there's the kicker, they
| actually have licenses and whatnot to the music, which I'm
| sure includes a clause for statistical / data-driven
| analysis.
|
| I wonder if that license extends to deriving a ML model from
| it. I know there's some ML models out there already that can
| produce music based on a prompt, but that'll be limited to
| what music the authors have available to them. Spotify (and
| Apple Music, and the others) have millions upon millions of
| tracks available to them.
| d4mi3n wrote:
| I suspect the RIAA might even be favor of supporting
| recommendation services--after all, matching people to music
| they like promotes music sales. RIAA members have a strongly
| vested interest in selling more music.
| mayhemchaos wrote:
| Our decision to cancel this project has zero to do with the
| RIAA. All the reasons are in the blog post -- it is that
| simple.
| at-fates-hands wrote:
| The hardest thing you can do when you're knee deep in a
| project is to pull the plug. I've been there a few times and
| it takes a lot of courage to do, especially when other people
| have put a ton of time and energy into it.
|
| I'm sorry it didn't work out.
|
| Is there a possible pivot to something else non-music
| related, or adapting your technology to another industry?
| sinuhe69 wrote:
| I wonder what kind of data do they have. Why did they say the
| quality of the data is not good enough for DL? Without concrete
| information, no one can provide meaningful feedback for them.
| Hamuko wrote:
| You can download their data (much like you can download the
| MusicBrainz database).
|
| https://acousticbrainz.org/download
| ricardobeat wrote:
| This page explains the data format:
| https://acousticbrainz.org/data
|
| Looks like they stored computed outputs from the "Essentia"
| tool, and the values are not accurate, so training a model on
| top of that will render equally innacurate results.
|
| To improve it you're [pun not intended] essentially starting
| from scratch.
| diibv wrote:
| It is worth noting that AcousticBrainz was based on an open-
| source audio analysis library Essentia [1] that has been
| gradually improving since the launch of the AcousticBrainz
| project. It now has better ML models, with higher accuracy and
| generalization, than those used in AcousticBrainz in the pre-deep
| learning age of handcrafted signal processing features and SVMs.
|
| See [2] for the list of models currently available. It includes
| updated ML models for the classifiers used in AcousticBrainz and
| many new models (e.g., a 400-style music classifier, models for
| arousal-valence emotion recognition, and feature embeddings that
| can be used for audio similarity or training new classifiers).
|
| [1] https://essentia.upf.edu/ [2]
| https://essentia.upf.edu/models.html
| tclancy wrote:
| This is unfortunate. A decade or so ago I had a beta for
| something IIRC was called MusicMatch which was just an executable
| that could create iTunes playlists based off characteristics of a
| song in iTunes and it felt like magic. Then Apple bought them,
| supposedly built it into iTunes but it seemed to simply disappear
| (DarkSky is experiencing this same phenomenon right now). Does
| anyone have a recommendation for things like this? Starting a
| radio station off a song in Spotify works well enough, but it
| feels like it just selects "People who listen to X also listened
| to Y" whereas MusicMatch worked off a fingerprint technique that
| was more serendipitous.
| piperswe wrote:
| Plex can do acoustic analysis on your library and uses that
| analysis for its recommendation algorithm - I find it works
| pretty well.
| jdminhbg wrote:
| > Then Apple bought them, supposedly built it into iTunes but
| it seemed to simply disappear (DarkSky is experiencing this
| same phenomenon right now)
|
| The Weather apps in the latest versions of iOS, iPadOS, and
| macOS are very obviously re-styled versions of DarkSky. They
| even rebranded and released the API.
| gfody wrote:
| this should be a cautionary tale for being careful how you model
| facts vs inferences, if you cant tell them apart your data is
| ruined
| TheOtherHobbes wrote:
| Music.
|
| Is.
|
| Hard.
|
| Understanding it and building useful models of it are really
| really hard.
|
| Even building a functional content-driven also-bot is hard,
| although you can always solve that problem by cheating and
| aggregating playlist preferences.
|
| Supposedly simple concepts like 'track BPM' just don't work
| reliably in the real world. (What's the tempo of a recording of a
| symphony, or even just a folk album that wasn't recorded to a
| click track? Or an EDM track with subtle tempo shifts - which
| quite a few tracks have?)
|
| If they'd known more about music when they started they'd have
| understood this.
| AndyKelley wrote:
| Will acoustID continue to be viable? https://acoustid.org/
| nix23 wrote:
| Wait what??? I used MusicBrainz for all my Audio and it works
| great!
| Hamuko wrote:
| AcousticBrainz isn't MusicBrainz.
| nix23 wrote:
| Ah true, they just worked together, thanks for the
| clarification.
| akino_germany wrote:
| Similar data is available from Spotify via their API:
| https://developer.spotify.com/documentation/web-api/referenc...
| adamzochowski wrote:
| This is based on old Echonest scanner. Echonest was bought out
| by spotify. Echonest had same issue as AcousticBrains, their
| first scanner sucked and mid way had to ask people to rescan
| using a new version.
|
| Echonest used to provide source code, but after being bought by
| spotify, they stopped updating their code, disabled fingerprint
| server, etc: https://github.com/spotify/echoprint-codegen
| mynegation wrote:
| It is unfortunate and it looks like it was a very interesting
| project, but I have so many questions. How come they looked at
| the data just now when they "collected enough" and not along the
| way. Quality of data and mis-labeling issues is something that
| would be apparent right away, no? What kind of "data quality" we
| are talking about? Is it raw waveform data? This has the ceiling
| of 44 KHz and pretty much everything is very close to it. Quality
| of existing labeling?
| arbitrage wrote:
| > Quality of data and mis-labeling issues is something that
| would be apparent right away, no?
|
| You are correct, those issues were apparent right from the
| beginning, and they never really got better. The acoustic
| fingerprinting worked sorta OK for very popular albums -- but
| even for that, it was never 100% accurate. It never worked well
| for live performances, imports, classical music, jazz, or jam
| bands.
|
| Using this software always required a lot of manual
| intervention, which at least for me negates the whole point of
| using it in the first place.
| nerdponx wrote:
| > We finally got around to doing this recenty, and realised
| that the data simply isn't of high enough quality to be useful
| for much at all.
|
| I suspect that they just didn't have the right people on the
| team, or those people were busy and haven't gotten around to
| it.
|
| This is sad, but it's an important cautionary tale for
| businesses that plan to collect data first, and later hire a
| data person to turn that data into gold. Many, many businesses
| in the 2010s probably failed because they thought they could do
| this.
|
| The truth is that you need data people involved from the
| beginning, and continuously throughout the project, in order to
| monitor and evaluate the data being collected, building proof-
| of-concept models along the way, and to adjust the data
| collection process as problems are discovered and new
| techniques are developed.
| wpietri wrote:
| Oh god yes. The "collect data" was such an "underpants
| gnomes" business model thing. Just thinking about it gives me
| flashbacks to industry events where insufferable people
| droned at me about this.
|
| And yes, another vote here for early proofs of concept and
| rigorous testing. Not everybody gets what really matters in
| data quality, but the surest way to find out is to try to
| build something and see if it really works.
___________________________________________________________________
(page generated 2023-01-19 23:01 UTC)