Nice approach. I'm the developer of https://www.blisshq.com/ and other projects. The key insight there is that managing large collections is easier with rules applied continuously, rather than thoughts-in-the-moment "I'll run a batch command with a tagger".
You could have the rules encoded in your config (or the DB) and apply those. For example, someone may want a "portable mirror" of their gold library, but that means transcoding AND chopping the embedded album art to size (or maybe their player has max resolutions).
I guess having the DB means you could potentially version control the metadata changes?
Oh very nice, bliss looks pretty slick. I see musefs as being kind of an unglamorous middle man layer that ultimately just has one job, separate the metadata handling from the backing files.
Since the DB is the source of truth you can absolutely version your metadata. The portable mirror use case is definitely one that people run into, but I see musefs more enabling that than handling it directly. It’d be really easy to handle transcoding and art handling just pointing at a musefs mount or by having a plugin for whatever you’re using handle it.
I definitely agree that continuous automation is the way to go, personally I set it up differently though. Whenever I have a new album in my music folder it gets detected and then beets imports it and museFS autoscans it. That handles 99% of cases silently in the background and the other 1% are the ones I’d want surfaced to make a judgement call on anyway.
That's what I mean by continuous automation, using rules (constraints) to dictate how the library is managed (in your case, what beets does, e.g. auto download art of the given size and format, constrain the genre etc etc).
At the back end of the 90s I did some work with self-organizing maps - collapsing multi dimensional space into a 2D map. I think it's interesting as a data visualisation approach. This looks similar but I'm totally out of date with where SOM went.
When you say 128d space, what are the data that are represented, exactly?
Hello! I find these fascinating too. The 128 dimensions I mentioned are learned by the NN, they are latent space and their actual meaning is ultimately unknown. The number itself is somewhat arbitrary as it was selected optimizing memory usage vs data coverage so that the track embeddings could sit on an A100.
Not only do A&A offer good speeds, their customer service is exceptional. There are chat rooms on which engineers and enthusiasts reside and if you phone them you get an engineer.
Short two wires on one end, go to the other end with a multimeter and test for continuity between those two wires on your lowest resistance range. If it is infinity then no connection. If it is some low number then there is a connection
Are you referring to this project specifically, or in general, how to create a PCB? If you want to create this project, you would need to download my exports in GitHub releases and send them off to a fab house, like JLCPCB. If in want to create a PCB in general, it's more similar to a connect the dots game, I used KiCad for this project
OP could use something like mp3fs to make the transition from MP3 to FLAC less painful where MP3s are still required - it's a lazy mirror of a FLAC collection to MP3.
(Disclaimer: this relates to the commercial project I run, but it is directly answering the parent)
It's not self hosted, more a middle-ground between rented Spotify and self hosted data sovereignty, but this is what we do at https://asti.ga . You store your music in some Internet-accessible storage, such as any S3 compatible endpoint, and Astiga connects and streams your library (and provides offline etc etc). AMA.
