PS Audio Music Server In The Pipeline?

Intelligent playlist building is a dreadfully difficult thing to do. Broken down to its basics, the idea is that the system should take into consideration what you have been listening to, and deduce from that what you might want to listen to next. This raises several difficult conundrums (conundra???).

  1. How long into the past should it consider your listening tastes? Just the current session? The whole week? All time? What you want to listen to tends to change with your moods and other factors. I may spend one day playing Jazz and the next day playing Classical. Do I want my intelligent system to play me a heady mix of Jazz and Classical for all time thereafter? I may buy a new album from Arcade Fire and play it heavily for a week or so. Do I want my system to start interspersing a heavy dose of old Arcade Fire tracks from then on?

  2. What is it about an individual track that the system should latch onto? If I play Eleanor Rigby, is it because I like The Beatles, or '60’s music, or ballads … or cellos? Mostly, the system can only associate the track to its metadata. I only know if it is '60’s music because the YEAR item in the metadata is set to 1966. Or it might have been set to 2002 because that is when that particular cut was remastered. Or 2008 because that was when the CD was re-issued. Or 1986 because someone, somewhere, entered some bad data. All are entirely possible. The system should probably consider GENRE as the most important metadata attribute, but for reasons too complex to address here, GENRE is typically the LEAST reliable metadata attribute out there, by a long way.

  3. Even if the system has some idea about what the listener likes to listen to, most listeners generally don’t want to hear the same stuff over and over again - they want the intelligent playlist to offer up something new. So it can only do that based on a combination of the attributes it already has of the tracks it has established that the user likes to hear, and the accessible attributes of the wider collection of available tracks that the user hasn’t yet heard. In particular, if those tracks come from TIDAL, or wherever, all it can go on is the limited metadata that might be available for that collection, or that might be usable to search that collection. That metadata may be (and in fact almost certainly will be) less comprehensive than the metadata available for your own personal library (particularly if that metadata has been collated by Octave).

  4. What do we do when the system offers up a track that we didn’t want to hear? We can in principle hit a “No, thanks” button to feed back info to the system that it can hopefully learn from. But, really, you would have to feed a lot more info back to the system in order to make the feedback effective. For example, did you mean “I never want to hear that track again”? Or did you mean “I don’t want a track with explicit lyrics while my neighbor’s kids are here”? Maybe you meant “I like that track, but I don’t want to hear it right now”? Or even “This track is too quiet and there’s a lot of background noise right now”. The distinctions are quite important.

Paul McGowan sent me an interesting link a while back, but I can’t find it any more. It was a reporter from something like the New York Times, and she was setting out her reaction to Spotify, if I recall correctly. Her entire viewpoint was based on what songs the system played for her after she had started off by playing some particular track. Two things jumped out. First, she kind of expected the system to read her mind, and make complex decisions on her behalf based upon almost no useful input data at all. Second, you could easily see where ten different people could have ten entirely different sets of expectations as to what kind of playlist they hoped the system would have delivered, given the same set of initial conditions.

An intelligent system would have to observe your listening habits for something like 5 years, after which time it could maybe be expected to make accurate decisions based on an assumption that your tastes have not evolved during that time.

1 Like

An excellent summary of the issues. This may explain why my experiences with playlist generation have been so negative.

I would find much more interesting the presentation of a significant number of recordings in which I may be interested based upon what I am currently listening to. This could be based upon all sorts of things: tempo, key, structure, time period, style, composer, etc. Not random, but broad reaching. This would likely tempt me to do some exploring of music I have not already heard.

(Given that conundrum is not Latin, we can make it plural in any way we would like. I enjoy your suggestion of conundra.)

1 Like

OE dictionary states it is conundrums. I believe conundra would be the Latin plural but as Elk said, it is not a Latin word.

1 Like

I will sleep better knowing that :wink:

1 Like

An interesting article on AI and machine learning by Jaren Lanier, widely credited as the inventor of virtual reality:

https://www.edge.org/conversation/jaron_lanier-the-myth-of-ai

He is an excellent musician. I had the chance to hang out as his house many moons ago. The walls had dozens of instruments from around the world.

What constitutes “intelligent”? To Lanier’s point we attribute intelligence to something that is merely pattern recognition, the machinations of a data base making correlations. There is no A/B testing possible here to determine what constitutes the success of one set of suggestions from another.

I did not think to look it up, silly me. Thanks!

Just here to try and help :wink: I probably abandon as many posts as I submit, as I like to read them through and if I’m not left thinking it was helpful I end up abandoning it.

That looks like an interesting read. I’ll be sure to set some time aside to go through it.

Mere pattern recognition can’t be sufficient to define “intelligence”. I think we need to add the capacity for deductive and inductive logic, and probably (although, maddeningly enough, I can’t come up with a non-theological justification for that) self-awareness. Either way, though, pattern recognition is fundamental to the way we as humans perceive the world.

I would also add abduction—the ability to make creative leaps based on analogous thinking. For a great read on this topic see Alexander’s The Jazz of Physics.

The ability to make and decipher meaning—let alone write algorithms—is far beyond pattern recognition. Meaning is contextual and multidimensional. Recently, an Uber in self-driving mode ran over a pedestrian. Some prefer machines taking over all aspects of human life. I’m not one of them.

LOL!! Don’t you just love spell checker :slight_smile:

I do hate spellcheck—again not very smart. But, for once, this is the word as I intended :slight_smile:

https://plato.stanford.edu/entries/abduction/

Also known as “inference to the best explanation”

I learn something new every day. So now I know of at least three separate meanings for that word :wink:

Very well stated. And this brings us full circle as to why music is so important, and also so personal.

I have 0 interest in “intelligent playlists” (Pandora’s artist radio is as close as I need) and I would prefer to see that time invested in improved user interfaces (and library management); what exists today sucks. The best I’ve used is Squeezepad/LMS, and it had much room for improvement.

1 Like

I thought the con-un-drum was part of the inner ear under the eardrum that convinces us the $$$ for that extra widget for the stereo is worth it? LIke the word implies, it ‘cons’ us into thinking the monetary outlay is acceptable . . . .

1 Like

No, I think you’ll find that word is “wife” …

SotM is releasing an audiophile Switch and it’s going to have Fiber optical input.

Fiber optical input would be a great feature for the server.

If you’re interested in a much better read on these topics, dig into this:

and this:

these are the two books that made me study philosophy, later sound engineering and become an audiophile :slight_smile:

1 Like

Once a DVD is written into the Kaleidescape ecosystem it becomes “region free” and you don’t need the physical disc stored in a disc vault. You can store it in a shoe box if you want (that is, for “grandfathered” systems like mine). For BR playback the physical disc needs to be present in the system somewhere (disc vault or player).

Thanks. It certainly gets complicated.

Actually they make the complicated simple. Whenever I read Roon, MQA, NAS, DSD, FLAC, JRiver etc etc in posts here, and all the difficulties people have with everything, I just quiver with dread. I have no difficulties at all. Bit for bit perfect copy of CD via coax to DAC. It works every time, nothing to think about, and it sounds great.

I’m just itching to get my Magico M3 speakers so I can hear how good a CD can sound. 10 week wait.