Improvement to lyrics function

So what about being able to edit lyrics? I constantly find errors and it’s very irritating that I can’t fix them.
I find that Genius usually gets the lyrics right. Maybe add that as a source to the lyrics?

Because of license issues, they probably don’t want people to edit or save them.

If this is the case then it’s probably best to implement a better source than LyricFind because the amount of wrong lyrics I find is rediculous

I also would greatly appreciate an improvement in lyrics. This specific point was made in this thread, and then Roon closed and locked the thread:

The OP in that thread summed up what Roon should do. This is precisely how Deezer shows lyrics. Big bold lines flow down the screen matching the music.

Roon is very crude by comparison. It is just a single page of text, and if you aren’t carefully following along you need to search through the page to find where you are and then scroll the page up, like using a word document.

Roon, you can do better than that.

Brian, you made that statement in January 2016. That was over two years ago.

When will it happen?

Not sure. We’re in a pretty different place than we were in January 2016, and our thinking has evolved on this quite a bit. I’ve written about this topic extensively before, but this is the gist:

One of the patterns that is repeating itself with this idea is–it’s more efficient to have machines make improvements than humans. Our user base is not millions–and the disappointing results that Musicbrainz/Discogs are getting in terms of coverage, completeness, consistency are not encouraging given that they are free and open to the public. I don’t think we would do as well as they have if we tried to copy that collaborative-editing based approach.

There is still a lot of ripe ground for automated cleanup, and we have several projects like that underway now. There also may be other useful applications for user-generated-content that bring more value per time invested to everyone than corrections/edits.

Back in 2016, our knowledge of Machine Learning + Data Science was very basic. It wasn’t obvious how to solve some of the problems that we’re facing in this area in an automatic fashion. Crowdsourcing seemed like a convenient escape hatch to solve what was then a nebulous problem, but it didn’t stand up to real scrutiny when we later pitted it against other more automated solutions.

Now that we are more educated and have more resources, it’s clear that there’s more efficient approaches available, and that both our time and our members’ time should be valued appropriately. I’d like to see a more balanced approach to crowdsourcing that has the machines doing the boring work and the humans doing the creative work.

2 Likes

Candidly, what does all of that mean? It strikes me a lot of rhetoric. Please explain in a simple, straightforward way.

Will the change occur, or not? If it will, when will it?

If not, why not? Is it not possible using automation?

Why can Deezer do it, and you cannot?

If it is not possible, I still greatly appreciate having lyrics in their current form, and urge you to at least retain that feature. Even if it is only a page of text that you must scroll through like a word document.

But something is better than nothing, and having lyrics in the current form is definitely a net-positive.

Thanks in advance for any additional clarification.

I was talking about crowdsourcing metadata edits/corrections–the thing you quoted me talking about in 2016–not lyrics.

That’s a complicated cost/benefit analysis between spending a lot of time building and administering a collective metadata editing system + investing in automatic improvements to our metadata library, not something that breaks down to simple explanation well.

There is also an aspect of that cost/benefit that takes into account how our members will use their time contributing. Should that time be used to make metadata corrections? To make playlists/collections of albums? To fill in gaps in the artwork library? To translate Roon into different languages? To aggregate an internet radio directory? … lots of possibilities, but same pool of effort.

My statement about the “balanced approach” is basically saying: we don’t want to waste humans’ time correcting things that machines could be correcting, and we probably won’t invest in human-driven metadata correction until we’ve run out of productive avenues with the machines.

I have already confirmed that time-coded lyrics are coming in other threads. We’ve secured the data…so there are no barriers there other than getting the work done at this point. I’m not going to make a time estimate for this publicly–we don’t do that in general. Too easy for exact plans to change.