My point of view would be that there are a number of large projects that they want to deliver but they need root and branch rearchitecture and as no timescales have been given to them I will enjoy each increment in functionality and feedback to improve what I see.
Brian’s explanation made it clear that the cloud association of the radio feature was not a business driven decision, but a technical one — they could not build a better local-only radio feature.
So there is no basis for questioning Roon’s interest in local libraries, for cases that do not involve such cloud-scale machine learning issues.
4 Likes
And that’s where the market is going, online and streaming so of course roon will go there. But the roon guys themselves strike me as big collection guys so won’t lose sight of that need to.
Oh, please do. It’d be great to have a fact-check on this.
2 Likes
Given the amount of resources they decided to spend on doing so, of course.
None of this is new; people have been getting Ph.D.s in music analysis and smart playlist algorithms for quite a few years now.
And as I mentioned above, the fact that you need large amounts of tagged ground truth to build machine learning models does not preclude them from being used with smaller amounts of data. There’s more to this story, though that “more” may be quite innocent and technical.
More generally, the fact that Roon will not provide a roadmap or timeline, and have actually ghosted features that in the ancient past were due to arrive “real soon now”, makes speculation about what’s next more attractive to those of us who have decided to each year reconsider our subscription fee. But it’s only speculation. Who can foretell the future?
And they are all cloud-scale, aren’t they?
Assuming a smiley there :-). Wow, how many posts would one have to read? Too many, I think.
Anders, with respect, what are you saying here? Yes, big data is needed to build the models; not to use them.
Nor will those who sensibly sign out of their TIDAL/ Qobuz account when they cease to pay for the service.
1 Like
New Nucleus & Roon user…
When scrolling for albums to play if you select and view the content then decide that is not the one you want it returns you back to the same stage you were searching alphabetically (essential for a lot of albums). If Roon does throw you back to A if you are scrolling P it is easy to use the slide bar to get back to P…
If you do the same thing in Playlists it always reverts to the 1st playlist (or sort criteria you have selected) and not where you were scrolling. If you have a lot of playlists it is frustrating to say the least.
Am I missing something or is this not “fixable”
Thanks
I was reacting to your full statement, Given the amount of resources they decided to spend on doing so, of course. None of this is new; people have been getting Ph.D.s in music analysis and smart playlist algorithms for quite a few years now.
Sounded as if you were disparaging the research they did in building the engine.
Wrt using the cloud-scale learning on smaller-scale data, yes, in some cases that works (like in my Haitian Creole example). How well that could work in this case is an interesting question: since your library is three or four orders of magnitude smaller than the cloud services’ I think that just running the engine, trying to play a result and finding it’s not available, trying the next one, would lead to a pretty bad experience given the latency of a cloud play attempt (999 failures per success). Some have suggested that, obviously not practical.
How well the algorithm can be adapted to local-only we don’t know, don’t know anything of the data representation.
But in any case, they didn’t build that. And not because they were unaware or uninterested in the issue, as @brian explained. I don’t think either of us is qualified to tell Brian that the technology they spent a year building can easily be adapted to another process.
They had strong demand for the ability of radio to go outside the local library (in these pages), found technical issues wrt implementation, and built an excellent solution for that.
2 Likes
Re: Point 2 by @crenca
This just happened to me.I was trying to see the values of a filter. Tried to scroll down using my laptop touchpad and ended up increasing the gain to distortion inducing levels.
The previous interface was far easier to use.
2 Likes
Oh, no. Sorry you read it that way. Should have put those two in different paragraphs.
But I was making the point that business interests are always “in the room”, as you know. They influence how much time you take, which people you allocate, which people you hire, how much you can pay them, how important various features seem to the management, etc. So saying that something is a “technical decision” really has attached to something like “given the various business decisions we made which affect this piece of the technology”.
Let me say one thing right away: I prefer to have a local collection and even so I’ve bought hundreds of albums at qobuz over the last few years I will not sign my Roon installation into my qobuz (non-streaming) account. So in a way it disappoints me a little that a new feature got released and I cannot make any use of it. But as @Chrislayeruk hinted: the model’s hopefully still in training – who knows, maybe in a year Roon feels more confident to also let local libraries profit from this Radio feature? Until then I resort to human curated radio - it still exists, Radio Paradise being just one fine example.
From what I remember Roonlabs never went so far as to really say something will come on day x - except maybe bug fixes. They quite cleverly - and, given the circumstances they operate in, in my opinion rightly so - avoid commitments of this kind.
Big Data & AI can! 
Wasn’t this the promise?
More seriously, there’s another thing around this “can only do it in the cloud” approach which I find disturbing and if it is indeed a hint into which direction Roonlabs intends to go I may waste a lifetime (subscription) on waiting for improvements I’d like to see. Right now, the improved updated search doesn’t work very well on a local only library and it seems to suffer from (temporary?) limits imposed on it which - my guess - originate from cloud service / performance restrictions and huge data sets I don’t have anything to do with. That gives me bad feelings. But I have the advantage that I don’t need to reconsider my subscription. If it ends, I couldn’t care less - either Roonlabs or me’s gone by then. 
Agreed. Brian even said that if there’s no connection to an external library, 1.6 will revert to the 1.5 implementation of radio. This is my case (no external libraries… for now 
) and last night my impression was that radio behaved the same as in 1.5. Even when pressing the thumbs down, you don’t get the popup asking you why (which I think is a feature of the new algorithm).
The Queue is the only place to go to see Up Next. I think Up Next should be available elsewhere as well. The Queue is a fine place to have it, and I understand the logic that Up Next is related to the Queue. However when Radio is on the Queue is not really the Queue as it is with a predetermined list of songs.
If you intend just to let Roon Radio fly you, but you want to see what’s up next, here is what you have to look at:
Maybe just maybe a little wasted space?
I would really like to see Up Next on the Now Playing screen as well, at least as an option.
4 Likes
I thought there was also a pop-up - well - popping up with the up next info and voting thumbs. Somewhere in the lower right corner?
Yes like 30 seconds before the song ends. It’s a cool feature. But you have to wait for it, can’t go somewhere and have that on the screen, plus it doesn’t have the album thumbnail with it.
Not disagreeing with you at all–a few things to add–
Definitely, and there’s an even worse problem with it that may not be as obvious. Even if you’re willing to spin the radio algorithm enough times to generate a radio stream, the 1/100 tracks that you’ll actually “hit” in the library are not well-distributed.
Lets say Radio builds a pool of tracks that spans 500 albums for a given seed. People mostly collect albums. You have 5 of those albums. Now radio is picking out of a 5 album shuffle really inefficiently.
Cue complaints about it being “boring”, “scattered” or “repetitive”, and “stopping randomly after a few hours”.
There’s no adapting…it would be a different algorithm, that is not dissimilar from what “library radio” is currently doing–building some sort of content similarity model within the Roon Core, then performing some weighted random picking based on a distance metric in that model, then applying business rules to keep things “on the rails”, “not repetitive”, etc.
Could we apply 20/20 hindsight and re-write library radio using some of what we learned from Roon Radio? Sure, but it would be a bad use of our effort. Lots of time and an ultra-long testing/iteration cycle to make something that already works OK work slightly better.
The “old” algorithm isn’t actually old–we released it 13 months ago. Sure, a total overhaul could make some modest improvements, but there is no new functionality to deliver there. It would make the same promises as the old feature and work in the same situations. Boring.
Roon Radio was worth serious effort both because it shares infrastructure with future discovery/exploration features, and because it adds a whole new dimension of functionality to the product. Sinking a bunch of resources into squeezing the next 5-10% improvement out of the library algorithm when we just overhauled it slightly over a year ago would be a big waste. Especially now, since a steep majority of our users have a better alternative–Roon Radio.
Also–it’s worth noting–we have been open about the plans for Roon Radio for a long time. It is and has always been on the same roadmap as library radio, with the plan to release in phases. Roon Radio is the final delivery from that plan that I laid out last year.
6 Likes
OK now you have my attention…
2 Likes