Fired up by some tagging and bookmarking discussions I’ve been having with @Paulb and @James_I in another thread, I thought it might be good to open a fresh thread to see if we can pool ideas for further development and to see what consensus of opinion there is regarding further development to these aspects? Ambivalence, strong views, lack of understanding, etc?
Currently I have been going great guns using tags for creation of my own genres and subgenres, and highlighting “good” tracks, and “NEW” stuff, etc.
The first question is why i feel the need to allocate my own Genres by using Tags… well, frankly because the Roon system for genre allocation is quite a fiddle to edit and I’d just like to be able to remove all Genres allocated by Roon and start again. There’s no way to create our own Genres and to find specific sub-genres to allocate to objects is difficult and there’s quite a bit crossover between subgenres using the system of allocation supplied by Roon themselves.
Re tagging:
The conversations I’ve had with James and Paul have hightlighted a common need for AND logis to be applicable to Tags. This would mean multifold more options for combining tags. It would make a much more elegant system of tagging. eg NEW AND GOOD AND GENRE could be used to combine 3 tags instead of having to assign a new tag for each combination of tag required. So in future, we could assign global tags for “Good” OR “New” to multiple objects and then combine these with Genre Tags or Mood Tags to get for example, all “Good” “New” “Blues” “Guitar” tracks which are for “late night listening”, All “Good” “Classical” “Concertos” from “Baroque” period.
Other requirements would appear to be a need to offer an option to switch off “intelligent” tagging which will display objects which are related sideways to specific tags - food for discussion.
That’ll do to get the discussion going
What’re your thoughts and how do you think Roon could implement Tagging, Bookmarking and Genre allocation better?
2 Likes
James_I
(The truth is out there but not necessarily here)
2
This is an important conversation. A lot of the ability to customize the Roon listening experience is supposed to be built into Tags, Bookmarks, and Genres. In some areas, Roon is already reasonably powerful, but in others, it is still pretty basic.
To summarize my other posts on were I think Tags need to go:
—Ability to import file-embedded metadata into Roon Tags. As discussed here:
—Multi-level tag hierarchy to allow drill down. Basically some method of organizing Tags other than flat on a single display page. Right now you can Tag Tags and you can use bookmarks, but these are kind of workarounds and requires continual manual organizational maintenance. An example is I have a set of Tags based on recommendations from friends. I don’t want to see this on the same page as my Tags used for administrative purposes like “Lossy Compression” and “Corrupt Files.” So I would want to have a page that shows only the top level Tag categories and then be able to drill down into a display of the specific category of Tag I am browsing.
—Tag sort logic as discussed here. This is a critical one as the workarounds are severe and limited in function:
—Ability to assign custom graphics to a Tag rather than the mosaic of objects within the Tag.
There is another elephant in the room when it comes to Tags. Paulb’s post does the best job of articulating this:
I have experienced this confusing disconnect as well, and I do not have an easy answer for this one. I think it arises out of the very innovative feature that Tags can be applied to almost any object within Roon - a track, an album, an artist, a playlist, another Tag, etc. In other apps, tags are inherently at the file level, which is usually a track. Even if the software, like Foobar, is able to display something so that it appears the tag applies to the album or artist, this is only because that same text string appears in the same field in each file tag.
In Roon, it is so much more sophisticated, but it is a kind of a web of relationships. I understand, this is the basis of Roon. But displaying in a simple, understandable fashion what tags are in a track when in album view, or displaying what tags apply to an album when looking at a track list, etc., is to me inherently difficult. You almost need a 3D display like in the bridge of a Rebel cruiser in Star Wars. I mean, there should be a way to understand what Tags have been applied to objects RELATED to the object you are viewing, but are not applied directly to that object, and having it function a little like a family tree. Anders brings up some similar ideas here, but related to Genres:
This I feel is super challenging. And tying browsing too tightly to Tags could lead to user errors and frustration – for example when I start to browse based on a Tag and my expectation is that only Tagged objects will be displayed, I might click down a given path, then leave, come back later, and not understand why all of my music appears not to be on a given screen. Roon almost needs to display my browsing path so that I can understand how I got to the subset so as to avoid me thinking there is a problem with the library. Not an easy one to fix.
OK, I have more, but this post is long enough. Thanks for considering!
To be honest, I don’t normally use tags, bookmarks or the genre filter. All I usually need is the search option (-> magnifying glass icon) when I look for a specific artist or album. I always find what I’m looking for — within a matter of seconds.
How do you see the Roon system retrieving the “rich metadata” that it advertises? I agree with most everything you’ve said. But the “free range chicken” model doesn’t comport well with the DBs upon which Roon relies.
The promise and pitfall of fuzzy logic systems is that in fixing one problem, they can create others. All such systems should ascribe to the Hippocratic Oath: “do no harm.”
1 Like
James_I
(The truth is out there but not necessarily here)
5
I don’t view Tags as a tool to aid in searching, per se. That is not to say it would be improper to use them that way - i.e. as a kind of mnemonic device.
To me, Tags are a way to supplement Roon’s playback/radio/shuffle function; a way to quickly build, in effect, a dynamic queue that matches your specific wishes at the moment. I think I laid it out best here:
2 Likes
James_I
(The truth is out there but not necessarily here)
6
I see it as somewhat separate from Tags and Bookmarks. Not that it could not be combined. But I really prefer to have different modes of listening depending on various factors like active listening versus working with music on, exploring a genre versus having happy hour with my wife, which requires a fairly restricted set of artists, etc.
So I view Tags as a means to construct my own Radio or playlist when Roon’s won’t do, using the logic and functions that Tags offer, and more is needed.
This is not to be critical of Roon Radio or Shuffle or anything like that. But no matter what, I categorically feel that there is no way that the Roon developers, or any developers for that matter, could EVER create an algorithm that doesn’t require me to have applied some level of Tags or other categorization to my music. Not unless I hired that team personally, I guess. It’s not being critical of what has been done, it is simply to say that I am a roll-your-own type of guy at times. I do use Roon Radio, I use shuffle all the time, and I also play based on Roon genres. But I do this as changes of pace. Unless Roon reads my mood and my mind, the Tags need to be there to play exactly what I want to hear on a Friday night or Sunday morning.
A Radio function with a whole bunch of sliders, switches, and levers would be pretty close to what I want without tagging. But Roon has already said that complex inner-gear settings are not in the cards for Radio. And I understand why - not everyone wants to get in there and tweak everything and they don’t think it would be a good product for the general masses. So that means, though, that I am rolling my own inside the Roon eco-system. It’s not a rebellion, it is leveraging this powerful system, that just needs a little more horsepower to be perfect for the task.
Yes I don’t disagree, and the sophistication of being able to apply Tags to any object has dramatically increased the challenges for the Roon team in displaying the relationships between Tagged objects and their parents or children. I don’t have a good answer. I only know there is great potential for confusion and I would like to see some sort of option to display those relationships while browsing.
Yes, when I know the name of artist / album the search option works ok.
But what I (and others) want to achieve with tagging (for me tracks, not artists or albums) is a set of songs with a word. Like taxonomy, a classification of a song to describe the mood, sphere or otherwise and to use in the future.
This I cannot achieve with the magnifying glass.
I think here is some of the problem, too many entities to tag.
I never use the shuffle or radio function. I don’t create “mood”-based or “genre”-based playlists, either. In fact, I don’t usually listen to individual tracks, as I very much prefer to enjoy albums in their entirety (i.e. whenever I have enough time for that). To each his own.
James, On the playlist front, I think iTunes’ Smart Playlist is the gold standard. Its the only thing I miss since migrating to Roon.
Here’s one idea I had: create a “random symphony program” a day. Maintain several playlists, auto-updated, one for each Musical Period. Then using those six or so lists, create a new lists that randomly picks, say, one from each list.
Take the same idea, but divvy the music by dates and create interesting contrasts of evolving music through time. Or by different composers, or different forms… You get the concept.
Here is the one and only window for creating a playlist:
Comparison operator options are “contains”,“does not contain”, “is”, “is not”, “begins with”, and “ends with”. Comprehensive enough, but it gets better.
Here’s the list of “tags” (to use Roon nomenclature) which can be used instead of Artist"
[Software may shrink true size; the bottom-most is “soft album artist”]
As impressive as the sheer number of tags/things one can play with (about 50!), the real kicker is “Playlist”. Yes, you can use playlists to build other playlists, allowing one to, say, build a playlist that’s 25% Piano Jazz, 25% Vocal Jazz, 25% Ragtime and so on. Proportional playlists, that automatically update with DB changes. Wow!
My point isn’t a lust for iTunes. In many ways it’s low-rent (no HQ stuff), and it underperforms badly in some areas (editing). But if I were building a way to get at my music, this is what I would choose. Very powerful, but can be used very simply if desired.
Agree. That’s why I have little enthusiasm for some of the scenarios for Boolean logic. Complex Boolean searches are not how I use Roon, and I think that is common. Plenty of research shows Boolean is a weak tool (except for programmers with years of education and training). In fact, that’s a very general observation: being good for programmers is not an indication that a tool is good for consumers.
Instead, we have seen the awesome power of the relevance-based search techniques on the internet. No Boolean in Google or Amazon.
I think it is important to recognize this, because otherwise those of us with such computer science backgrounds would know exactly how to improve Roon, b6 emulating the logic constructs of programming.
But if we focus on the scenarios you touch on here, guiding shuffle and radio, we might find more interesting and useful techniques.
(Although I won’t contribute because I never do shuffle or radio.)
Hmm. I think Google contradicts that. No controls, no tags, but it is very good at finding relevant results.
The challenge is, the Google algorithms are based on processing vast amounts of data with vast amounts of computer power. Won’t fit in a NUC. One question is, can we use cloud processing and big data to solve this? I think this is what Roon is trying to do, based on something @Brian wrote a while back. Maybe what we are seeing is a learning curve. Google 1.0 wasn’t perfect.
James_I
(The truth is out there but not necessarily here)
12
Wow. I have not used iTunes for anything but managing my iPhone or just literally uploading music to an iPhone. Never listened to it as a desktop music manager. its lack of flexibility regarding formats, especially on PC, has always warded me off.
But that is impressive. I had no idea. I would love for Roon to implement something like that, especially if it supported custom categories like our own Tags. It would be beautiful.
I suspect there are a few patents there, and so Roon would have to work around that…,.
James_I
(The truth is out there but not necessarily here)
13
Well, I guess it is theoretically possible but it seems like a huge amount of programming to be the Google of music and then we would have to be very careful in formulating our searches to get what we want, relative to the iTunes style Smart Playlist that gives you specific options as to how to search/create the list of objects.
I mean, sure, if Roon can give me a good playlist when I type in “Playlist with a mix of mellow country-rock from the 1990s and downbeat Electronica from the 2000s, except no Illbient, and mix in country-punk from the 1980s as 1/20th of the list” then I am all in.
BUT, to me, that is a release in 2022, and until then, I’d like to be able to leverage my tags to get to 95% of the same result…
You would know more than I. But I thought Google also uses some human intervention to create equivalences as well? I mean, doesn’t Google manually go in and say “hors d’oeuvres = appetizers” ???
You might try it. I find it very enjoyable to have my PC suggest music for me right out of my own library, mixed with stuff I know well.
Not typically, because such a technique is not scalable, which means it would not have an appreciable impact on a $100 Billion business.
One experience that shows the value of the new techniques, and the weakness of what we did in the previous century: experts have been work on machine translation of human languages for 40 years, using formal grammars, vocabularies and thesauruses, Boolean logic and all the power of classical computer science. But human language has always been “30 years away” and looked like it would always remain so.
But then they tried using machine learning, processing the vast data volume of the internet.
The result: when Haiti had the earthquake catastrophe a few years ago, Microsoft Research built a bidirectional translator between Haitian Creole and English, in 4 days and 7 hours.
EDIT: this is why the New York Times recently wrote that Silicon Valley is recruiting machine learning talent and data scientists by offering NFL level salaries. (For the non-Americans, NFL is the National Football League, American style football.)
Exactly. Anyone who thinks humans fixing mistakes or cleaning data entry by entry is the answer is living in a world that no longer exists.
There is a lot of grumbling about the quality of Rovi’s data on this site, and often an implicit assumption that humans should be fixing it–but to my eye, that data source looks like exactly what you get when you try to employ Humans to do that kind of task.
Some problems are easy for computers and humans. Easy for humans, hard for computers. Easy for computers, hard for humans. Hard for both. When you leverage computers more, you get more of the easy-for-computers problems solved. This shapes products. Google generally doesn’t do things that require armies of highly accurate people–they just don’t even try to build those products.
Doing more automated stuff isn’t a silver bullet for all problems. It’s going to allow us to do a lot of things we aren’t doing at all now. It’s going to improve the quality of some things we are doing. It probably is not going to enable us to conjure missing data from nowhere. The waterline for the final experience goes up–but the problems that are hard for anyone to solve are going to remain hard.
Good point HWZ but perhaps we don’t need to make a distinction for the purpose of this discussion?
Actually, I am using tags for 3 different “objects” - Tracks, Artists AND Albums. I think we all like to listen to albums too
To all intents and purposes the Tagging and Bookmarking discussion need not differentiate between Albums or Tracks per se, imo?
Think of an album as a “super-track”
I am using tagging and bookmarking to make complete MiscoSD Card album sets by genre et al, for portable use, mostly. Also I have family members who can’t bear certain genres (Jazz for example) and so I need to filter offer such “abominations” for casual listening.
Everyone will have a different usage case or set of cases.
The goal is to give ideas to Roon to refine what we have in terms of tools to get the results we would like to achieve.
I suppose that depends on your definition of Humans Sure, data entry errors are expected but in the case of Rovi I get the sense they are exploiting their apparent monopoly position to deliver a mediocre product. Maybe I’m misunderstanding how the process to add an album’s data to Rovi’s database works, but if you are adding two saxophone players to the list of credits, one called Chris Potter and another Chris Porter, it merits a double take on the source of the credits. Maybe I’m old fashioned.
Maybe Rovi should employ some automation, at a minimum to flag exceptions like these. An algorithm could highlight close name matches that could be the result of a typo or it could detect that an artist credited with playing the saxophone has in the past only played drums. Maybe Rovi already does this? I doubt it.
What worries me is that the Roon team clearly sweats the details when it comes to this —admittedly niche— domain of music enjoyment while it’s not obvious that Rovi does. And, unless I’m mistaken, Roon can’t exist without Rovi.
