You say you dont think it’s an issue, but just in case…
It’s possible that Roon does know about these artists but your artist is not being matched with Roon’s. I have sometimes added a credit for a well known performer only for that person to come up with a blank bio and image.
The way around that is to ensure that the artist is already in your library beforehand, even if that means adding an album and deleting it later.
I put a couple of attempts by the industry at standardization above
a full treatise would look like War & Peace
Put 10 Classical Lovers in a room to develop a standard and you’ll get 11 opinions 
@Jez,
I think you are taking this out of context. I am just as frustrated as you about this specific navigation problem. But the discussion was not really about that. It just crossed into a use-case of particular interest to you. The point of my posts was to find out a bit more about how natural language processing can be applied to music navigation generally, not just this specific case.
To me music is not free form text, it is structured frequency waves. I imagine a future where search and library organization systems navigate the music itself not the crude text identifiers we attach to it. I want to be able to ask questions like “build me a modern composition oriented playlist without serial music that is predominantly female voice and that will enhance my current mood”.
If we are to avoid the dead-end of incomprehensible tagging standards, that will require a deep domain understanding of the structure of music itself, not the crude text identifiers we currently tag it with. I assume that we are a long way off from that, but I maybe wrong. I would like to find out. On the specific case of interest to you (and myself actually) I doubt that what I am talking about is necessary but TBH I have no idea what the technical challenges really are as I have no professional expertise. Someone must have mentioned this already. It’s not ideal but surely simply sorting the list by playing time would be a step forward?
Do we really need as flexible search engine ala Google to find a Piano Concerto , esp where the tags are probably there anyway
OOPs is this too logical for most ?
I am sure if you ask Roon will refund the balance of your year , It’s clearly not for you after your limited investigation into its features.
I am sure they don’ really want an unhappy user spoiling the usual good humour of this forum 
(Grumpy because of LockDown !!)
I opened another thread on the subject “Search”. Just another can of worms producing 90% arbitrary results. Very little consistancy.
This software is so bad - its hard to believe people fork out 700 USD for a live-time license…
After all of this, I will just love my good old trusted UPnP Control point even more. Free of charge btw.
Very good point. I find things instantly on my UPnP control point. Roon makes a science out of it, produces 586 matches, of which 550 are crap. Sigh…
Nope, the more I investigate the more inconsistencies I find.
Obviously I can’t go into detail about how the software works. Let’s stick to this: it can correlate data points from unstructured and from structured data pools and -lakes.
Roon seems to be opposed to structured searches, though they should be perfectly feasible and are easy to implement.
Their current search is supposed to be “context sensitive”, tough where the context is defined is anyone’s guess since the user cannot define it.
My client’s software can deduce context, if by no other means than a natural language question.
The extra data for classical music might come from public domain resources of which there are plenty for classical. This could be correlated with the actual data points in Roon’s database.
Intelligent stand-off markup if you will. Without the need to predefine the markup relations.
I don’t know. Do we? We don’t seem to get the results we’d like as things stand now.
What is your “Clients Software” ? proprietary ?
No, you didn’t. Manual album identification isn’t the same as Roon’s search function…
I appreciate you cannot go into details but even in general terms you must correlate between two completely different data lakes. From what you are saying one is some kind of free form text lake from public sources but the the other is a completely different frequency wave lake (music). How in general terms is that going to be done?
One obvious way is to attach text identifiers to the frequency wave lake (tag the frequency wave lake so to speak) and then there are two text lakes for natural language correlation. That is in effect what we do at the moment. But my understanding is that the problem is not really the correlation, it is the tagging in the first place. Maybe natural language processing can give advantages but it doesn’t really change a general approach that is just not scaling. International standards are not really being agreed and there appears to be few commercial drivers to significantly improve the data quality we have at the moment.
It is interesting though if you are able to be a little bit more specific, just in general terms why the current meta-data industry, such as it is would benefit.
Yes.(10 char)
No! Just something that manages the simple things. Like a search for a Beethoven Piano Concerto 3 shouldn’t return Mozart Piano Concerto 21. And if you select the composition, then you should get instances of the composition not excerpts.
Also, searching is a different issue from how people choose to organise their music. People should surely be allowed to organise their music how they see fit. If you buy a chest of drawers from IKEA you aren’t forced to keep your socks in the top drawer ordered alphabetically by colour. But that is exactly what Roon does with its fixed views. Very restrictive.
Not something I would normally search for , it came back with all 5 but nothing else , not unreasonable
Doesn’t matter where the data comes from, as long as it can be interpreted. As for the how of interpretation: sorry, classified.
The problem with tags is that the result depends on the quality, pertinence and consistency of the tags. Correlation information does not need to come from tags or such. You need to link the question to the actual data to be retrieved. This is the correlation part.
My client’s approach is not data-centric, it’s interpretation-centric. As long as the end result of the interpretation part can be mapped to where the relevant data is available for retrieval, it will work.
My client’s software is able to read the available metadata text (forgive me the inaccuracy, I’m talking about the artist information, bio’s and such) and can use this to find relevant information. Like Roon’s discovery feature, but infinitely more accurate and relevant. It can also draw on available information outside of Roon or your local library, provided it is told where this can be found and of course provided copyright is respected.
Let go of what you know about data retrieval. This software works in a totally different way.
But linked to what. Other text? I certainly can see the case for a better domain specific (interpretation based) google or wiki. But a music player must organise music not text. Can your clients software analyse a WAV file and determine if it follows a classical concerto form, is pitched at A440 Hz, predominately uses a major or minor or other scale or whether the bbm is slower/average/faster for the performance ? Or is it expecting that detail to be available in a text file somewhere? It’s not so much a question of the quality of this information (in any form) anymore. The direction seems to be to produce less and less of it in the first place.
I’m considering a retrieval scenario based on what is available as retrievable data (music) on one hand and information about the retrievable data on the other hand.
The retrievable data must of course be identified in some way or another in a classical way (excuse pun), i.e. file references as we know them (title, performer, composer, track or part,…).
The retrieval process in this kind of application would be initiated by a) getting relevant information from the query (i.e. all recordings of Mozart’s Kleine Nachtmuzik") and b) inferring additional information from the results of this interpretation.
Provided there is enough available data (e.g. publishing catalogs, public domain information) the query could be correlated to available items in the library, so you would get the items requested, including divertimento no 15, serenade voor strijkkwartet, etc., but not other divertimento’s etc.
It’s a question of using the information at hand as opposed to what happens now, which is casting a wide net and dumping everything on the user.
It’s just a suggestion, BTW. I’m not saying my client is the miracle worker who will save the day or that Roon should use their software - which isn’t free, they run a business after all. I’m just convinced that this product could do away with a lot of the frustration without user intervention like having to normalize tags etc.
It’s an interesting idea. Not sure where you will get the inference rules from. I guess that’s the trick of this? Inference rules for each application domain. I am aware of strong commercial drivers in some domains, medical, legal, public policy come to mind, as I vaguely work in this area myself. I’m sure there are others. Not so sure about Classical music though! Idagio for example just used a brute force approach of hiring a large bunch of musicologists. I assume they digitized all that professional knowledge in some way. Not sure if there is anything like that in the public domain though. But good luck with it. It will be great to see the next generation of music players start to emerge.