top of page
Search
  • Steve

NAM Plugin version 0.7.6 is released

Updated: Dec 14, 2023

Update 2023-12-14: I've just released v0.7.7, which fixes a bug with this version that was causing crashes for some users. The download links have been updated.


After a long hiatus, I've finally got a new version of the NAM plugin for you! As always, you can download the installer from the users page.


What's new


Support for all DAW sample rates

This is the big one and I didn't want to make a release without delivering on this. Thanks for your patience!



Plugin while using an interface sample rate of 44.1 kHz and a model that expects a sample rate of 48 kHz.


Has this ever happened to you?


"WARNING: NAM model expects sample rate 48000"


No more! Now, the plugin takes care of the sample rate conversion internally so that the models operate as intended, and you hear what you expect.


NAM is tricky because, unlike classical modeling approaches, it expects the samples it's processing to come in at the same sample rate as the data it was trained on. With this release, I've added resampling to the plugin so that you can use your favorite sample rate in your DAW, and the NAM model will still see the audio at the sample rate that it wants.


This also applies to NAM's IR loader--previously, there was a bug where IRs [sometimes] weren't getting resampled correctly as well.


In summary, the NAM plugin now sounds the same no matter what sample rate your DAW is running at.


Support for touch screens

Nothing much to say here...folks using NAM on touch screen devices should be able to manipulate it using that touch screen!


Added a DC blocker

Another surprising part about the machine learning methods that NAM models use is that they don't "know" that there's anything special about the input/output "zero"--like the fact that an amp should output silence if the input is silent. This means that the same prediction error for normal playing applies to silence--NAM gets it wrong by a tiny bit (just how much depends on how the model was trained, and each one is different). Unfortunately, this means that NAM seems to have a low "noise floor" that can mess with downstream plugins like noise gates and loudness meters.


To fix this, I've added a low-frequency high-pass filter ("DC blocker") that pulls the output (slowly) back to zero, recovering that "silence in, silence out" property that most audio gear has.


Normalization is on by default

NAM's plugin has a "normalize" switch that makes sure that different models' output levels are adjusted to sound (approximately) the same loudness as each other. This switch used to be off by default because I figured that people might be more interested in having the model behave exactly as the reamp it was based on. However, the much more common use case today is that users didn't make their own models and might be switching between different models made by different people. To help them smooth out the volume jumps and avoid surprises, I've put this on by default.




Enjoy!

3,501 views

Recent Posts

See All

NAM Plugin version 0.7.9 is released

I've released a patch for the NAM snapshot plugin. Download it on the users page. The full changelog can be read on GitHub. Nothing much to say except this release focuses on a bug fix for the resampl

Towards a (good) CPU-efficient NAM

One of the most common problems people have with NAM is that the standard WaveNet architecture (the default in the simplified trainers and therefore the most common) is too compute-hungry to work in a

Comentários


bottom of page