St. Rock AMPERIUM LIVE

[2dor]

Someone was asked for it :)

Turned out great \m/

 

[2dor]

Any chance of a quick tutorial on how to do it locally? Cheers

It would be great to have an easier way to train locally. Been trying to get it up & running on my PC but can't get around some of the stuff the trainer's spitting out; I know for a fact that the files are the right sample rate (44.1K) and in the correct location.

 

[2dor]

Figured out I didn't get the file paths correct for my PC for the "in_file" and "out_file" variables; got the trainer running now but it's so sloooooooow... looks like it's running on the CPU instead of the GPU (I've an RTX 4070 I use for NAM training).

 

[40DayWarningK]

Ease of use of such features like amp model training like this is really crucial for wider adoption. This desperately needs to be made into an easy to use app. Devs need to look at it from a newbie's perspective. This is partly why I believe NMM hub doesn't have much user participation.

 

[2dor]

Ease of use of such features like amp model training like this is really crucial for wider adoption. This desperately needs to be made into an easy to use app. Devs need to look at it from a newbie's perspective. This is partly why I believe NMM hub doesn't have much user participation.

I agree. I haven't been able to figure it out properly by using the NAM training sequence ar 44.1KHz and a reamped signal; I've been training NAM models locally since February last year probably.

 

[deLuther]

 

[deLuther]

I agree. I haven't been able to figure it out properly by using the NAM training sequence ar 44.1KHz and a reamped signal; I've been training NAM models locally since February last year probably.

NAM training sequence requires resampling to 44.1 for both test and reamped signals. Also start of training part requires change of default training parameters (i.e. offset).

 

[2dor]

NAM training sequence requires resampling to 44.1 for both test and reamped signals. Also start of training part requires change of default training parameters (i.e. offset).

I did measure the samples between the peak in both signals and supplied them.
Will try the Proteus signal today just for the kicks.

 

[deLuther]

NAM @48 has around 1.17 second offset (exact value in samples specified somewhere in code, don`t remember exact value).
Proteus has 1 second and ToneX has 5 second.

And both files should be properly aligned as with proper reamping.
Automatic latency determination in NAM or ToneX not always give you correct result, especially with high gain.

 

[deLuther]

Last of 3 new "Jose" amps :)

 

[2dor]

So I'm in the process of training my first Amperium NNM model in the Colab cloud using the Proteus singnal for reamping.

It took a bit of trial / error to get the right sample offset in there. I had to put in a negative value for the delaySamples (caught onto this from when I was training NAM models, the automatic delay was always in the -32 samples range when I had 48K reamps).

ESR seems to be going down with the epochs which is good but I hate Colab. Would love to have a tutorial on how to properly train locally using a GPU.

I was able to run the Jupyter notebook locally from Anaconda but the epoch count was slow AF compared to what I've seen NAM does.

 

[deLuther]