On 2023-01-13, Andy Farnell wrote:
It it therefore vigilant of Sampo to pick up on the anomalous source
address - but it could have been handled more elegantly Sampo, perhaps
with a simple query: "Are you really from Apple?"
Quite the thing, and not untrue.
Though if I asked that, obviously "they" would say "yes", with no means
for me of ascertaining if it was actually so. In fact I could easily
feign to be an Apple-er. Any day of the week, hour or second.
Pretty much the only way I can discern whether someone on-list is "nice"
is by format. Whether they recognize the age-old conventions of internet
mail.
For instance in this mail the format/morality hasn't been followed.
Because the email address is in all caps.
BTW these research posts look awesome, and if I were a younger man I'd
be all over it.
Yes. Indeed. I *am* all over them, evenas I prove an amateur.
The opportunities for mixing new machine learning advances with audio
DSP are tremendous, as I discovered in my time at HyperSentience
(Mogees).
Any links towards that? Because, frankly, I've yet to see a *single*
audio-DSP application which crucially relies on deep learning.
Perhaps we can turn the discussion to that subject.
Let's.
Back then we were using Kohonen SEMs and simple K-means to do "context
awareness" on Mel spectrums.
Even a link to that? I'm well privy to Kohonen's work, especially since
one of my friends even befriended the guy. And, yes, once he did some
inverse Volterra work in his teens, something like Yamaha, maybe Onkyo,
maybe Matsushita, shut him down via an international patent suit.
What advances does anyone think the latest ML methods might bring to
figuring out if you're on a bus, at a cafe, in a sports hall?
Because that has amazing applications to auto-adjust headphones etc.
I actually don't know. I'm kind of stupid, in that I learnt this stuff
so that it's mostly linear. Sure, I can understand nonlinear control
theory and whatnot, to a degree. But utilizing deep learning as the
highest order nonlinear learning gambit here...I actually don't even
know what it should be used for in audio- or musicdsp. For maxiimal
effect.
Okay, fine, since I tend towards ambisonics theory, maybe, just *maybe*,
machine and deep learning could be utililized for spotting point sources
there, even their distance, and quite certainly recurrent neural
networks in time would be well suited towards such a task. In particular
if you do transform layers from earlier theory, all the way to
homomorphic signal processing/cepstral. Applied all over the sphere, as
ambisonic quadrature does. Whew, the possibility of doing low weight
superresolution *is* there.
Yet I don't think that will be done by people who spell out their Apple
addresses in all caps.
--
Sampo Syreeni, aka decoy - de...@iki.fi, https://urldefense.proofpoint.com/v2/url?u=http-3A__decoy.iki.fi_front&d=DwIBAg&c=009klHSCxuh5AI1vNQzSO0KGjl4nbi2Q0M1QLJX9BeE&r=TRvFbpof3kTa2q5hdjI2hccynPix7hNL2n0I6DmlDy0&m=o3XG6z0I8m3tDAEjgwj-W8mUSlqMWUpMV5DvwuP-W1O7Y1q2uUGIAGM94blCK5Zk&s=Tz5R_lTQxMiSR1mj1DhSInAmCD6ccsHoi6J1nQVm5t4&e=
+358-40-3751464, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2