Machine Learning Feature

[mccy]

I just looked up for that machine learning feature and found it under utilities in the MTurbocomp... yeah! Right now I was testing the Cenozoix Compressor and I thought why not trying a ML ... after more the 10000 rounds (took aboult 30minutes) I'm at a max difference of -135db which should be more than enough!? for the VariMu type...
It is still working, so no listening comparison so far... I'm really curious...

EDIT: I was so curious that I interrupted and found that I used a wrong file for "ideal"... so next try

Who is interested in exchanging results? Wasn't there the melda online repository or something called like that!?!?!?!

[Andernik81@gmail.com]

+1

[Frankie.T]

I tested it a lot, and i came with a concusion the way is made cannot understand how a compressor works.
I wrote more about that some times ago, forgive me now if i'm lazy to re-write it, but to write it fast and badly it needs a dedicated set of audio with different characteristics to "study" first the shape differences and behaviors.

I studied few compressors myself and the way some works is particular with different behaviors between peaks, sustain and relative volume along other stuff.

So to reproduce them you need to feed ML with the needed data and tests

[Klisyaeron]

I just looked up for that machine learning feature and found it under utilities in the MTurbocomp... yeah! Right now I was testing the Cenozoix Compressor and I thought why not trying a ML ... after more the 10000 rounds (took aboult 30minutes) I'm at a max difference of -135db which should be more than enough!? for the VariMu type. I notice that now machine learning can do very smart things. While experimenting, I came across the idea of ​​how to Build an AI-powered Recommendation System which seemed interesting to me. If ML can help people get personalized help with movies or products on websites, then it could also offer to automatically adapt compression settings in real time.
It is still working, so no listening comparison so far... I'm really curious...

EDIT: I was so curious that I interrupted and found that I used a wrong file for "ideal"... so next try

Who is interested in exchanging results? Wasn't there the melda online repository or something called like that!?!?!?!

does this require intensive use of the GPU or the CPU? I recently upgraded my graphics card, but my processor is still pretty old

[MeldaProduction]

It's purely CPU, implementing this on GPU would be near impossible. But it's fast enough. The main thing is to design good test signals. Contrary to my original assumptions I ended up using some real word audio and not any scientific test stuff.