top of page
Search
  • Steve

NAM Trainer v0.8.1 is released

Updated: Feb 11

Edit 2024-02-11: Released version 0.8.1 with bug fixes.


I've just released a new version of the NAM trainer. As usual, the new version is used automatically in the online trainer,and folks training locally on their own computers can update their install from the terminal by typing

pip install --upgrade neural-amp-modeler

Full release notes are on GitHub.

What's new?


Improved Colab trainer UX

I've simplified the appearance of the trainer on Colab: All of the options are exposed in the nice form UI, and training has been simplified down to a single click!

Simplified Colab notebook. Drag your files in, hit the "Play" button, and come back in 8 minutes!


Support for Proteus reamps

There are a few different projects aside from NAM that make neural models of gear. One of those is GuitarML, whose Proteus has a very similar workflow to NAM's. If you've reamped your gear using the Proteus_Capture.wav file from Proteus's capture utility, you can now use those files in NAM's simplified trainers (Colab & local GUI trainers) to make a NAM from the same data!


Better support for non-48k training

In addition to Proteus's files that are supported, the CLI trainer now easily supports reamp data with sample rates other than 48k. You're still responsible for making sure that your model architecture is reasonably-designed for your data! E.g. if you use the standard WaveNet without modification on 96k data, it will see the same number of samples into the past--which is half as long in real time!


Breaking changes

Bumping the minor version from 0.7 to 0.8 means that a few things no longer work. This is all on purpose and helps me keep the repository clean and focused.


  • Removed nam.data.REQUIRED_RATE. If you used this constant in your code, then you'll need to do something else.

  • Removed bin/run.py. This isn't all that useful anymore given the C++ DSP code and the plugin. If you still want it, you're free to grab the script from an older version and hang onto it yourself, but I'm not supporting it anymore.

  • Removed bin/export/main.py. This isn't needed anymore because the training script makes the .nam file at the end automatically now!

  • Removed nam.models.parametric and nam.data.ParametricDataset. This is the code for parametric models and data sets. If you were using it, you can easily re-add it to your work by using the model and data set registry functionality that I added in last versions. However, allows me to make fewer assumptions about how NAM gets used and gives downstream developers more control over how they choose to go about modeling.


Enjoy!

1,001 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

Comments


bottom of page