Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[baldo]

I have a music library of almost 100K files, give or take a few K!!!

Some of these files/ tracks I have never played and may have never heard. 

Any ideas on how I can go about exploring my own music collection?  I could compile a playlist of tracks never played. That is one way. 

I could by the Genre, by Era by the date that I added the files. 

Is there any kind of AI software that can go through my collection and perhaps predict what I would like.  Maybe by looking at how often certain tracks are played or used in playlists, and then finding similar tracks.

I would welcome any thoughts and insights on this.

[JimH]

Play Doctor.

[baldo]

ok. How do I use Play Doctor?

[eve]

I have a music library of almost 100K files, give or take a few K!!!

Some of these files/ tracks I have never played and may have never heard. 

Any ideas on how I can go about exploring my own music collection?  I could compile a playlist of tracks never played. That is one way. 

I could by the Genre, by Era by the date that I added the files. 

Is there any kind of AI software that can go through my collection and perhaps predict what I would like.  Maybe by looking at how often certain tracks are played or used in playlists, and then finding similar tracks.

I would welcome any thoughts and insights on this.

Very complex. I actually do these kind of recommender systems for myself on the side.

One problem if you're trying to tailor a model to 'you' is that, you still need user data from other people. An easy way to think of this is basically, you summarize 'how you score' various albums, and compare that to how alot of people score various albums. You find a person that has similar tastes, and you can use their 'score' for other things to fill in the gaps for 'you', thus with a variable degree of accuracy, predicting what you'd score that album you haven't scored. You can do similar things with user playlists as an input, 'predicting' a track that would fit nicely next to a given set of tracks.

That's about the most basic explainer of how a rudimentary recommendation model works I can give, it's not very good.
 

[JimH]

It's on the wiki.

[eve]

It's on the wiki.

Out of curiosity, how do you guys handle Play Doctor's choices? Any insights?

[JimH]

Take a look at the settings available in MC's options.

You can also choose the degree of variety you want.  And whether to use local or Cloudplay or both.

Then it starts with the seed file or files you choose. 

Then it learns as you skip or delete.

Marko did a good job here:  https://wiki.jriver.com/index.php/Play_Doctor

[baldo]

Ok, I found Playdoctor but its does not look particularly useful beyond picking some random tracks from my library!

How can I set it so that it picks files from say a particular Genre. Or from a particular Era, 1970s, 1980s etc ( I have all my music tagged into and Era).

I looked at the Rules and set just one rule, Era contains 1970s

and I get no tracks coming up.  Which is clearly wrong.

I looked at the Wiki, I'm sorry but that was not helpful.

[HPBEME]

Click on the Search DJ button right next to PlayDoctor text box.  That will bring up a dialog box where you can enter genre among a few other options.

[eve]

Take a look at the settings available in MC's options.

You can also choose the degree of variety you want.  And whether to use local or Cloudplay or both.

Then it starts with the seed file or files you choose. 

Then it learns as you skip or delete.

Marko did a good job here:  https://wiki.jriver.com/index.php/Play_Doctor

Was this a response to me or the OP?

[baldo]

There is not Search DJ button next to Play Doctor.   I have the words "Options" underlined and when I click on that I get the chose to set some rules. I set one rule, the Era to be 1970s. And I get There are no files.

What we need is some kind of AI. should not be too difficult to programme. Just use the parameter of how often files are played and then Dj software like Mixed in Key to find similar songs.

In fact the only reason I have a Spotify and Apple Subscription is to discover new music. THE ONLY REASON!!!

[JimH]

There is not Search DJ button next to Play Doctor.   

Are you using an old version?

It's on the Playing Now page when playback is stopped.

[HPBEME]

I don't remember for sure when Search DJ was added, but it's been around for at least a few major versions at this point.

[dtc]

The best system for selecting related music is Pandora. It is based on setting a couple of hundred of parameters based on the characteristics of each track. That is a huge undertaking and none of the other streaming services is nearly as sophisticated in creating a "radio station" of related music, simply because they do not have the level of analysis that Pandora has.

The takeaway from that is that any intelligent selection system is only as good as the metadata that it has to work with.

Within MC, a search on multiple existing metadata fields, and then shuffling the results is probably as good as anything else. If you want to save those searches, then playlists or smartlists are probably as easy as anything.

As far as I can tell, Search DJ is basically a search of your metadata based on 4 metadata fields. It does not seem that there is much analysis beyond that being done. But it might be a place to start.

A really good AI system to do this would be really nice, but you need to do a lot of analysis to get really good relationships between tracks.

[marko]

I don't remember for sure when Search DJ was added, but it's been around for at least a few major versions at this point.

Added in build 31.0.51, added to Playing Now in 31.0.52

[baldo]

Ok, my version of MC does not have DJ Search.  And if all it does is search a few meta tags then that just not worth the hassle or cost of upgrading.

I have absolutely no doubt that there is an AI tool that can do this. Heck, Apple music, Spotify et al are probably already implementing it.  All of the Data already exists withing MC. Like last played, how often a track is played etc etc.  Maybe its not the skill set for the MC developers (No disrespect intended) and what we need is different paradigm.

Maybe the question to ask would be "how can I get access to ALL of the data that MC has for my library?" Once we what that we can then give it to an AI to sift through and find patterns and come up with suggestions.

[JimH]

Why don't you try Search DJ, Cloudplay, and Play Doctor?  The combination is pretty amazing.

[dtc]

Have you used the search wizard? You can set up any criteria you want based on the metadata. So, you can, for example, use 3 genres, a date range, a set of ratings, number of times played, etc. Then shuffle the result.  You can do the same thing with smartlists.

[eve]

I have absolutely no doubt that there is an AI tool that can do this. Heck, Apple music, Spotify et al are probably already implementing it.  All of the Data already exists withing MC. Like last played, how often a track is played etc etc.  Maybe its not the skill set for the MC developers (No disrespect intended) and what we need is different paradigm.

Yes, however you'd be mainly looking at YOU. The patterns for developing models, based upon things like listening data tend to require a relatively decent sample size of people.

Now, JRiver could collect this information, the problem is everyone's metadata is different, so the models would end up being ALOT dumber because a text string  like "The Beatles" as an artist,  only goes so far. They'd basically need to do what I do internally and resolve these things to more concrete "entities" For example, a 'track' is not just a track, it's a recording + composition (or compositions technically), with multiple participants (parties, or artists in more common parlance).

I'm pretty sure you could do some kind of CSV export from JRiver. I track all the playback activity my JRiver instances get (only for video) but it's external to JRiver (basically it's part of my wider, internal analytics setup)