I guess their issue might be that if people listen to white noise 8 hours while asleep every night and Spotify distributes their monthly fee weighted by time listened, then a lot of their fee will go to the white noise for sleep and thus the total fee distributed to their daytime music would be unexpectedly significantly reduced.
The easy fix would be to cap track time weight, although that is susceptible to an attack of splitting the white noise in many parts and publishing a playlist.
I guess a possibility could be to somehow compute "song variety" (i.e. entropy in a model where the decoder can generate noise/randomness) and weight by that, but not sure if the available lossy codecs are good enough to do that.
The entropy in white noise is maximal, if it's really white noise.
One sample of white noise is indistinguishable from another; so if Spotify can identify white noise, then they can dedup, i.e. serve the same sample for every request for white-noise track.
What I find annoying is that people are wasting bandwidth uploading and downloading an undifferentiated hiss. White noise is trivial to generate locally, without consuming any bandwidth.
[Edit] Real white noise has the same energy at every frequency; the total energy in white noise is effectively infinite. Practical "white noise" is low-pass-filtered, which means it's no longer real white noise.
I wonder if these samples are really pink noise, and Spotify is talking nonsense?
A finite-length sample of the most perfect white noise will not attain maximum entropy. You mention a lowpass filter -- that's accomplished through the sampling frequency. But your proposal is an effective lowpass filter, as any frequency longer than the sample length will clipped to exactly the sample length -- which you'll be able to hear as a distinct rhythm if it's a sub-audible frequency, or a tone if it's shorter.
Well, Spotify serves lossy compressed audio anyway. I don't know much about audio compression, but presumably it's first bandpass-filtered, and then filtered to remove psycho-acoutically unwanted audio components, and then compressed as a bitstream.
I agree with your interpretation, but I'd like to also offer another way to view this phenomenon that Spotify's higher execs also missed: Spotify is lucky and has found a way to monetize a feature that is readily and freely available elsewhere, and they should just be happy to get a bit of profit for it.
I would take the opposite approach and figure out how to maximize the return on the white noise without changing anything for the users; that is, focus on reducing the cost to deliver white noise, work on guidelines for the white noise presenters on how you're going to monetize this without disrupting the fad, etc.
From my perspective, I just can't see how trying to do something special with this fad does anything but immediately kill the fad. There are even FOSS white noise apps, and it won't take long for users to find a free alternative if Spotify messes with the recipe here. I sincerely doubt anyone is going to get Spotify exclusively for the white noise nor that white noise will somehow be a gateway into further Spotify use; I just don't see that the persons who want white noise would use that as an entry point into the service, it's the other way around, with current satisfied users finding out they can also use Spotify to get white noise.
Basically I see this as a happy accident for Spotify that will break if they try to press on it too much. They should treat this like a beneficial fad, and just figure out how to deliver it with the least resource cost, and just enjoy the extra revenue. I don't think it's really going to draw people in except if they play the "yeah, this is legit, we're just gonna get out of your way as much as possible here. enjoy our ads", and ride that money until it dries up.
I hate to say it but certain advertisers might pay more for ads that play while people are sleeping (ie placement during 8 hour tracks of white noise).
This is not a $38m-in-costs crisis for Spotify, this is a revenue opportunity!
I think this is why google really wants to control the brower. Once AI is cheap and on every device, the user has the ability to use it to filter all inbound. Trivial for a local LLM to strip hate speech from a twitter html regardless how much Elon wants you to force you to see it.
Most people don't even use an adblocker, despite how easy it is. I think you will have a hard time trying to get a substantial number of people to install a filter for opinions you/they don't like.
You may want to actually look at the adblock usage. It’s massive unless artificially disabled. Enough for Facebook and Youtube to fight it with escalating countermeasures
The easy fix would be to cap track time weight, although that is susceptible to an attack of splitting the white noise in many parts and publishing a playlist.
I guess a possibility could be to somehow compute "song variety" (i.e. entropy in a model where the decoder can generate noise/randomness) and weight by that, but not sure if the available lossy codecs are good enough to do that.