You can do both: click the “next” icon ( ->) for the song playing. Roon will ask you why you skipped. In the queue screen on the right, you can thumbs up or thumbs down the upcoming song. It will be queued with a thumbs up, skipped with a thumbs down.
Doesn’t mean Radio gets better though.
Go to the “Now Playing” screen, and you’ll see the thumbs there for the song that’s currently being played. To go to that screen click on the bottom of the screen, where the track/artist is listed.
You can do that by repetitively clicking the thumbs up button for the suggested Roon Radio track. That allows you build up a play queue of Roon suggested tracks.
What I mean is that I want to be able to see the queue that Roon has concocted. The screen on the right will only show you one track, initially only just as it’s about to play. I can only see the rest of the queue one item at a time, and if I want to browse it, I can only do so by manually cancelling each previous track by hitting the thumbs down button, which is no good if I happen to want to listen to it.
I think its an incorrect assumption that Roon has built a queue, rather than offering you choices one at a time. If it offered a queue, it would presumably be nearly infinite.
But if you are on the queue screen and just keeping clicking the thumbs up button a queue will be built as fast as you can click the thumb. At that point a queue is built and you can treat it like any other queue.
It may not be what you were expecting, but I don’t think there is a good alternative.
I missed that you could make the queue visible by clicking the thunbs up button, thanks. However, I assume that the the thumbs up button indicates that I like the track, which may or may not be the case.
Given that the queue is generated based on music I’ve played in the past, I still don’t see why it can’t just be displayed - at least the next 10 or 20 tracks.
Well, you’ve hit the nail on the head there. Nobody knows how Roon generates the next song. All we can do is speculate.
What I’ve learned so far during the 1.6 period:
It’s catering to the lowest common denominator … aka people who bought that also bought this. I don’t think it knows anything about the underlying music.
As far as training it is concerned I’d venture to say your thumbs up/down are aggregated into a pool so individually it means little - Valence is catering to the user community’s collective tastes rather than your own. Perhaps I’m wrong, but that’s how I understand it works.
The blurb on 1.6 promised us something quite different, though @evand:
The all-new Roon Radio
Roon has always been great at reconnecting you with the music you’ve collected over the years, but now we’ve completely reimagined how Roon creates your personalized radio experience. Designed from the ground up using our new machine learning technology, Roon Radio will automatically play a mix of your favorites, along with relevant selections from beyond your library.
Based on more than a year of research and development and several months of alpha testing, Roon Radio makes recommendations based not only your own music preferences but also those of other Roon subscribers with similar tastes in music.
As you listen and interact with Roon Radio over time, it learns more about what you like and helps you discover new music by curating great recommendations.
Roon Radio considers metadata-based relationships, listening history, overall popularity, and significance of content, to build a model that predicts the affinity between radio picks and your Roon profile. Internally, Roon Radio will also consider dozens of expert-opinion dimensions–the classical-ness of content, whether it is holiday music, and so on. It uses a variety of techniques to avoid irrelevant picks, excessive repetition of the same content, inappropriate/irrelevant content playback, etc.
Not much - if anything - of all this has been delivered. Above blurb mainly promises a personal Radio generation, aided by community sampling.
Catering to the lowest common denominator is the exact opposite of what the blurb announces.
Bloody hell! Elton John again! How do you get from Pink Floyd to Elton John!?
Does he have shares in Roon?
I think it means it’s going to take time to learn more about your preferences and how that intersects (or doesn’t) with the community’s tastes.
I think nearly a year should be enough time to “learn” at least a little. The result however, is a step backwards.
It’s not the lack of positive evolution that galls me though. It’s the bragging and not delivering that gets my goat.
I like the radio and it delivers for the sort of music I like. I don’t want a radio to only play songs I know and would predict so like some of the quirky picks.
I’d be hacked off to get Elton after Floyd also
I agree Roon Radio is now giving me tracks I don’t like and have no interest in. For example Chinese language covers of cheesy pop songs. Also I personally find the voting questions that come up every time I skip a song to be really annoying.
Similar experience (and feedback) on my end. Pre-1.7, I thought Roon Radio was one of the “killer features” of Roon. I was really impressed with the way it intelligently selected new tracks for after my current queue completed. It was a valuable tool for me in terms of “discovery”.
With 1.7, I’m finding a lot of value in their new “Recommended for You” column to the right… it seems like the brains/smarts that were once used to select next tracks for Roon Radio are now being applied over there. And I’m getting a lot of use out of that.
But last weekend I officially gave up on this newest version of Roon Radio… I wasn’t (personally) having any luck with it anymore, and I finally turned it off.
Roon only asks why you skipped a song if you don’t have RRadio set to play from your local library.
I concur. Something happened. I went from listening to Johnny Mathis Christmas album to Linda Ronstadt What’s New, about as far apart in genre as I can conceive
Well I’m glad I’m not the only one. For me it keeps playing Bruce Springsteen, and I never listen to him. I mean yesterday I played hours of early to mid Rolling Stones from my library and then comes Bruce Springsteen?? WTH? and then 3 more songs that had nothing to do with my listening habits, my library, or mid 60’s Rolling Stones. Guess I’m just gonna turn off radio as it’s not getting better nor learning. It used to be way better.
Maybe it depends on the music genre, or even specific sub genre that you want to hear at the time.
In general I find if I want the best out of it, I have to carefully choose a middle of the road track of the style I want to listen to. I also find it will vary a lot from one day to the next, some days a really good selection, others I can find myself skipping a lot of out of place tracks.
As for using choices made by lot of other people, I think that is reasonable starting point (it needs to start somewhere) if over time it will refine to your own choices.
In the past I have spent a lot of time doing deep programmatic analysis of some genres of EDM and TBH didn’t really get anywhere with it. My interest at the time was an experiment to see if it was practical to autogenerate a reasonable DJ set of a given style and progression. Even with key determination (for harmonic mixing) and quite a decent beat recognition machine it just couldn’t pick out some important key elements in common between tracks. even getting into chord progression analysis didn’t really help that much - maybe a team with far greater skills than I may have done better, but I wonder how much better. It doesnt help that even I often find it hard to say specifically why I might like one track and yet may dislike another that would probably analyse to be very similar and so in theory a good fit. I would suggest that in some way EDM is possibly among the easiest to analyse, so using crowd preferences seems like an understandable starting point.
Lots of people have tried to do this - I am yet to find anyone do it successfully according to my personal tastes - there are just too many complex influences. There is also the consideration that a lot of what we like may have nothing to do with the music at all, but may come from memories associated with hearing it in the past - a place/time etc. I personally listen to a lot of EDM, and yet, I also would skip most EDM because it is musically too simple, or the production quality isn’t so great, or lacks the energy I want or an overly processed vocal style or it just sounds like too much other overly commercial junk or whatever. There are different types of people when it comes to what it is about music that grabs them. Some people like music, some like vocal for eg and barely care about the music behind.
The whole thing is too complex I think for a machine to figure it out on its own without a lot of help and the best help there is probably the associations made by other people. I suspect the association being used at this time however are far too simple - ie too few levels. Just because some random person like track A and B and someone else like B and C, it doesn’t follow that a person who like A will also like C. I think it needs to be taken a few stages further to start to understand clusters of multi-level associations, but that is some CPU heavy analysis that goes up by some power of N selections - ie probably computationally impractical without throwing an entire google datacenter at the problem.
I don’t know much about AI, but that sounds right. The only stuff I know is what I see on YouTube videos of dudes trying to build AI’s to learn how to play a particular, simple video game. Just one dude, with just one desktop PC, can make a lot of good progress, but after many hours of tweaking and sometimes making the problems easier to solve. Music taste is so many orders of magnitude more complex. Also, I don’t think Roon is throwing actual AI at this project - the choices just don’t seem at all sophisticated.