Why unidentified compositions in unidentified albums?

Often an unidentified album will have a combination of identified and unidentified compositions. To me it is even more important to identify compositions in unidentified albums as it integrates an otherwise “orphan” album into my library.

Scenario 1
However, if these unidentified compositions are singletons (that is, I have no other instances in identified albums) then there does not seem to be anything I can do to force roon to recognise them. By definition these are singletons so there are no other composition instances I could merge them with. But I also find that changing the titles to match the canonical names on allmusic will not force an id either.

Scenario 2
The pattern I notice is that if I import an identified album that has compositions in common with my unidentified album then suddenly the unidentified compositions in the unidentified album are now identified.

It seems to me that roon is aware of the compositions in both scenarios but chooses only to make the relevant links in scenario 2. If roon has the composition information it should be making the identification and all relevant links in both scenarios. Has anyone else noticed this behavior? In many ways getting as many composition id’s as possible is even more important in scenario 1 than scenario 2 as it goes a long way to compensating for the lack of an album identification.

3 Likes

Unfortunately, Roon does not identify a composition unless it occurs in some identified album in your collection. I’m not sure of the technical reason, but my guess is that without an identified album, the Roon metadata reconciliation algorithm is not sufficiently sure they have a uniquely specified composition. Composition identification is tricky, especially for poorly catalogued works or works that have multiple versions (different instruments/orchestrations, revisions by the composer, publisher editions,…). At least with curated, identified album metadata, its compositions might be (but they aren’t always) accurate.

Don’t understand why,if you enter the composers
Manually,they stay hidden under a tab,and not
Under the song title
Why Roon does this in beyond me

1 Like

I can see that. But if roon is relying on identified albums to “make sure” of a composition identification there isn’t much basis in either logic or real world metadata experience.

An album being identified is no guarantee at all that the composition has been correctly identified. I have countless examples. It’s a frequent source of posts here. I would say that it is my most frequent editing task with identified albums. Hunting down opus and catalog numbers and canonical names. Normalising the spelling of composers, reordering the order of composers, re-assigning performers as composers. Roon frequently also gets composition assignments wrong in identified albums where there are multiple versions/orchestrations/transcriptions even with the core repertoire. For example the difference between Tchaikovsky ballets and suites. The French repertoire (particularly Debussy where dual L numbers are being used) is a complete mess. If anything I find pop/rock much worse. So this is not something effecting only more obscure and specialist material and already causes a significant manual editing overhead if it is important to you.

I would say that I was probably more meticulous with unidentified albums simply because here is no underlying automation I can rely on. Maybe there is scope here for roon to revisit the underlying logic of the composition identification logic in the light of experience with real world libraries and real world meta-data. If someone from roon can confirm that in this case roon is working as expected I can raise a feature request. I would also be interested to hear the reasoning as at the moment the logic defeats me.

Just to make the discussion a little more concrete. On this unidentified album of 10 Bach cantatas, roon identified 0. That is zero out of ten.

A couple of points.

  1. I have changed the titles to the allmusic cannonical forms. The chances of me getting a 100% match to rovi like this seems to be absolutely zero. Surely that is enough confidence for roon to make a match. Clearly it is no accident, no coincidence, beyond all reasonable doubt.

  2. This is my “travel” roon. A sub-set of my main library that I use on the road. As such I am deliberately being selective with what I take with me. No doubt these cantatas are identified in my “home” roon as I will have other cantatas in identified albums. So it is particularly annoying that in smaller libraries in this and other scenarios I am going to get a lot of unidentified compositions, despite the fact that I am prepared to edit canonical names.

Roon displays composers by default under the song/composition if you have a Classical or Jazz genre.

If you want to force roon to display composers for other genres you can change “settings/show composer credits” to “more” or “always”.

1 Like

I’ve just experimented with an album that I know Roon does not recognise to see if it makes associations to composers where one of the tracks is a known composition by another artist. It does - see final track:

Yep. That’s what I see as well, provided of course the composition has already been correctly identified in the other identified album :grinning:

In my scenario I have a much smaller “travel” roon about 15% of what I have at home. What this is meaning in practice is that by proportion I have much fewer identified albums and this is having a “multiplier” effect of much more unidentified compositions. The logic roon appears to be using is having the effect of making composition identification a function of the size of your library.

But TBH I just don’t get why an album identification necessarily means a correct composition attribution or even any attribution at all as roon seems to be assuming. My experience is that identified albums, depending on genre, often have no or few composition identifications and it is not uncommon for those that it does have to be wrong. That can lead to the very bizarre and difficult to correct scenario where the composition in the unidentified album has been correctly attributed but paired with an incorrectly attributed composition in the identified album. I have had this happen several times.

My starting point is rovi canonical names. I guess in their efforts to make them unique, particularly with classical they can end up looking very odd and unique indeed. If you have gone to a conscious effort to use them in your files then it seems to me roon should give the benefit of the doubt and attribute the composition whether it is in an unidentified or identified album.

I agree with your question about Rovi canonical names. I have several old classical boxsets that I’ve carefully tagged with WORK and PART tags following Allmusic.com, but the compositions are not identified unless they also occur in some identified album. At the very least, an option to identify compositions even if albums as a whole are not identified would be nice. Many of those older albums and boxsets are not identified because the classical record catalog from the 50s to the 80s is a mess with multiple regional issues, re-issues, compilations into boxsets, label name changes from industry consolidation,…

Totally agree with this.

in my experience, album identification has nothing to do with composition attribution.
i mean, i have a lot of both identified AND un-identified albums where:
-all tracks have a corresponding composition
-most tracks have a corresponding composition, bus some don’t
-only some tracks have a corresponding composition

it’s something i know has come out other times on the forum, something i have asked many times to the @staff and for which i have received NO ANSWER AT ALL.

and i bet no answer will come in this thread.

Hey Nicola, I share your frustration.

I think the best way to get attention to these matters is to just be as specific as possible. Many here are benefiting from your patience and the detail with which you regularly point out metadata issues. Thank you for that. Please continue.

I think on this particular thread what it is boiling down to is a concrete requirement that rovi canonical composition names should be honored whether they appear in identified or unidentified albums. This is a hopefully relatively simple and concrete requirement that will benefit many, whether they realize it or not. It’s also something concrete that roon can prioritize and asses the impact. So the best thing I can probably do is raise a feature request summarizing the comments here.

3 Likes

Hey Tony
actually i think the situation is quite complicated at the composition level. as far as i understand the whole topic, the main problem is… rovi database. simply, there are tracks that do not have a composition and a composer.
file tagging is useless… because file tags are always (?) interpreted as TRACKS CREDITS and not as COMPOSITION CREDITS (which are different… this has a logic, but the double level is quite cumbersome…)

so, the solution would be:
-if a track has not composition credits, use file tags to create them
-allow users to EDIT them. in this sense, i find roon very frustrating… no way to customize, no way to manually edit things when… things coming from roon are wrong (or nonsense)

but… i know (at least they say) that they are working in a new metadata engine, that should solve many metadata issues (doubled artists). and i hope that this will solve also this part of metadata inconsistencies.

I don’t think roon is ever going to open their entire model to user edits. This is only of interest to a very small handful (including myself) and at the end of the day it doesn’t fit into their mission. So although that would be nice for a few of us, its a bit unrealistic. But I do have the impression that roon are open to smaller changes where the benefits can be demonstrated.

A small pragmatic change is being proposed here.

At the moment canonical names are only recognized in identified albums. The idea would be that manual edits of canonical names in unidentified albums should be recognized as well.

Roon’s concern seems to be the possibility of false positives and they are using an album identification as a sort of a composition match “confirmation”. However, for several of us that is not making much sense as we see so many cases where identified albums have no composition attributions or even mistaken ones.

On the other hand a canonical name in an unidentified album is never going to be found there unless it was deliberately put there by a roon user. The chances of a false positive must be close to zero.

Taking a famous example, the rovi canonical name for “Also sprach Zarathustra” is:

Also sprach Zarathustra (Thus Spoke Zoroaster), tone poem for orchestra, Op. 30 (TrV 176)

I have never seen a formulation as unique and complex as that on the back of any CD I have. Rovi go out of their way to make these canonical names quite unique. Other examples, can be much more extreme. So, if I have deliberately gone to allmusic and cut and paste this into the tags of an unidentified album then roon should be pretty safe in assuming that this is indeed “Also sprach Zarathrustra”.

This seems to me a pragmatic and pretty safe change that (hopefully) is not that onerous to implement. But it would have a lot of benefits for those prepared to do the edits particularly with smaller libraries.

1 Like

hi tony. i don’t undestand. i have non-identified albums where tracks do have a composition… that should be the “canonical name” for the composition. so… i don’t know how roon associates a non identified album with compositions: i guess that there are cases where file tags TITLE and COMPOSER makes the system confident in associating to a composition in its database.

for example



but other tracks in the album are not associated to any composition.
accidentally, i have other tracks corresponding to that composition, but they are recognized as different. and no other track is matched to composition “capriccio diabolico”

another example:



here all tracks are associated to a composition. some of them are well known. but most of them are quite unknown: composer is benvenuto terzi (my grandfather…), and on allmusic this album is missing. i don’t know how comes that roon associates a composition, but it seems that they are not in the database.

and:



here all tracks are associated to a composition.
and i doubt that there exists another album with the same tracks that could work as a crossing match.

on the other hand, in identified albums with tracks non associated to a composition, i have many cases like this:


the album is identifies. but track 3 is not associated to any composition:

on the other hand, the composition DO exist in my library:


and it is in a different version of the same album by nina simone, identified.

so, the 2 scenarios you describe don’t seem to match what i see in my library. i don’t know.

one final point is this: if the album is both identified OR non-identified, there’s nothing you can do with the “singletons”. i mean, as far as i can see there’s no way to merge the singleton with the composition associated in another album. in fact to merge you need something to merge… in the case of the nina simone’s album for exampe i connot manually associate the singleton track with the identified composition. at least, i have found no way to do it.

I think we are talking at cross-purposes.

Whenever you enter a composer, roon will automatically make that “go to composition” link but it is not necessarily linking to a canonical composition which has an existence outside your library in rovi. In order to get that link you need the little CD counter link next to the track (which you don’t have in your David Russel example).

It is much easier to see the difference with classical than with pop because with classical, with a canonical identification you will get a lot of composition bio and other data. With “songs” this is much more hit and miss. Most of the time there is no bio and other metadata attached to the song.

I see this all the time. roon can link your compositions together because they have common composers and spelling but it hasn’t necessarily linked your compositions to a rovi instance outside your library. That is the sense in which we are talking about a composition identification here. What at least three of us have noticed is that roon will not make a composition id in this sense in an unidentified album unless it can cross-confirm with an instance in one of your identified albums.

1 Like

it’s not so. for example, in the david russel album all tracks have a composer tag (and work and part, like most of the tracks in my library). but the “go to composition” is created only for track 3.
adding a composer is not sufficient to have a composition, local to the library or present in rovi.
do you have a different behavior? because actually i have searched quite a lot in the forum, and i did not find any explanation.

Well, that surprises me. I wasn’t expecting everyone’s roon to behave differently. Whenever I add a composer roon automatically creates that link. Here is a typical example of an album I haven’t got round to tagging. It’s unidentified, there are no composers and no compositions.

If I add a composer, even a fictitious one, roon will automatically make that link but it is obviously not linking to anything canonical, in the sense of a composition that has an independent existence from my library. The “composer” is my wife’s pet name for our cat:

It’s obviously not possible for me to test every track in my library but I haven’t come across a counter example yet. But from what you are saying you are having an almost 100% opposite experience. That doesn’t make much sense to me so there must be a difference, maybe in the way in which we add a composer? All I do is write it directly into the files using mp3tag or within roon using the track editor. The result is always the same.

Tony, thank you so much.

and YES, THAT SURPRISES ME TOO!
we are experiencing totally different behaviors… and now i can understand you first post.

possibly, the different behavior can be tracked back to some library setting somewhere. but i don’t know where.
here, i can only tag someone in the @staff (@joel, @dylan) to have suggestions on the problem, or to redirect me for having support about this.

I believe Scenario 1 (and actually 2, too) work(s) as currently designed - you may remember this thread:

I don’t really know but my guess is the convenience effect of Scenario 2 is the result of locally performed recognition = no Roon “cloud processing” for items below the album level if the album isn’t ID’d.

Now if this design could or should be changed is another question. And I’m not sure if relying on what Rovi thinks the canonical name of a composition should be is future proof enough.

You have a good memory (I’m getting old :zombie:). Joel’s response doesn’t actually say that a condition of a composition identification is at least one corresponding album identification.

However , if you don’t have another instance of the composition in your library which has been identified by Roon (i.e. come from Roon metadata), Roon will not, as yet, “identify” a composition (against Roon metadata) from file tag metadata.

There is of course the argument that an album identification is an implicit condition of a “safe” composition id and from that perspective roon is working as designed. I accept that. It’s just that several here with real life libraries are simply not seeing a particularly strong correlation between an album identification and a “safe” composition identification. So from that perspective the question is what other criteria could/should be used to increase the “safety” of a composition identification?

One possibility is canonical names. They don’t necessarily have to be rovi, that’s just a pragmatic choice because for the foreseeable future those are the ones roon is aware of. What’s important is that they are unique, have been user edited and are extremely unlikely to be coincidentally false metadata. I use them all the time in identified albums where my experience is of relatively poor rates of composition identification compared with album identification. It would be nice to see the same functionality extended to unidentified albums without the restriction that an identified composition in an identified album already exists.