NAM vs NNM vs ANN comparison of neural network profiles of tube preamp

[@vaar5323]

Андрей, а где можно скачать этот профиль для Amperium? Звучит замечательно!

[@АндрейПолевой-я5и]

Ютуб не даёт возможности ссылки вставлять, поэтому надо заменить все запятые на точки ниже и @ на /: drive,google,com@drive@folders@10gcPgEJXZKUTJdSdqN-E7dqh89WKfnGR?usp=drive_link

[@АндрейПолевой-я5и]

Также вытащил на amperium hub, надеюсь, правильно вспомнил, что за файл использовал в видео

[@vaar5323]

@@АндрейПолевой-я5и ого, спасибо большое!

[@NO-TALK-GuitarPlugin]

To give you my feel I found NNM more convincing. NAM has something « granular » and more « explosive » that is not the case in preamp. Curious as on the paper NAM is supposed to be better. But ESR is not all we need ;)

[@АндрейПолевой-я5и]

ESR is just most popular one now, i also measure and look at mae, sometimes at mse, huber... Interesting, i can't hear here this granular explosive NAM sound. But for shure, most difference i feel when playing through this models, how it reacts...

[@exltd001]

@@АндрейПолевой-я5и

[@dietcoke8114]

i wanna ask about ann training, did you use the ori.wav file provided by sinco or did you use your own di track sound? i always get nan error during my training session. thanks in advance btw

[@АндрейПолевой-я5и]

@@dietcoke8114 here i used long training file - union of ori.wav with proteus, nam and tonex training files. You can try to use 1024batch with slow learning rate - it will train longer but can possibly help with nan. And if you get nan - close training window and run training again - it will start from last nit nan epoch and maybe will not go to nan

[@HMohr]

How can I get your captures for tank g?

[@АндрейПолевой-я5и]

Can't post links here, you can mail me to [e-mail] or search links in Facebook tank g groups by my name Андрей Полевой

[@NO-TALK-GuitarPlugin]

Nice but null testing would be a lot more easy to evaluate 😊

[@АндрейПолевой-я5и]

Yes, and loss level valuable for it) I think will do next video with null-test. But here i want to find out if there is a significant difference to ears.

[@АндрейПолевой-я5и]

Loss value (ESR metrics): NAM 0.011 NNM 0.038 ANN 0.071

[@NO-TALK-GuitarPlugin]

@@АндрейПолевой-я5иnice, but esr doesn’t reflect really what human ear can or can’t perceive (but esr is the tool I agree). What would be interesting would be null testing AND EQ Signal Range Null testing (haha inventing new concept here) to measure null testing within the human ear and specifically what we can discern into a mix for example 400Hz to 6-8kHz (there’s a lot of energy in the 20Hz to 200Hz but we don’t really care and perceive it in mixes. Your thoughts?

[@АндрейПолевой-я5и]

​@@NO-TALK-GuitarPlugin on my other null tests main differences was in freqs 5kHz, i think here will be the same. But on NAM here i used Fourier and MrSTFT losses too, so it must be better, will see)

[@NeZversSounds]

@@NO-TALK-GuitarPlugin human perception is flawed - measurement above feelings.