Add support for embedding NAM models and IR files in session data #595

damiangr · 2025-11-05T13:35:20Z

Description

This PR adds the ability to embed NAM model files (.nam) and impulse response files (.wav) directly into DAW session/project files, making sessions fully portable and self-contained.

Features

Automatic Embedding: When saving a session, NAM models and IR files are encoded and embedded in the plugin state
Automatic Extraction: When loading a session, embedded files are extracted to temporary files and loaded automatically
Backward Compatible: Sessions without embedded data continue to work with file paths as before
Cross-Platform: Designed to work on Windows, macOS, and Linux

Benefits

Share sessions without worrying about missing model/IR files
Old projects remain playable even if original files are moved/deleted
Team members can open sessions without manual file management

Implementation

Modified SerializeState() to embed file contents in session data
Modified UnserializeState() to detect and extract embedded files automatically
Added _StageModelFromData() and _StageIRFromData() helper methods for loading from memory
Maintain backward compatibility with file path references
Update AudioDSPTools submodule to include SIGFPE fix (depends on Fix SIGFPE crash - Initialize mRawAudioSampleRate and validate sample… AudioDSPTools#24)

PR Checklist

Did you format your code using format.bash?
Does the VST3 plugin pass all of the unit tests in the VST3PluginTestHost?
- Windows - Validated with pluginval (strictness level 5) - SUCCESS
- macOS - Unable to test: VMware Sonoma 14.7.2 + Xcode 16.2 environment has NanoVG/Metal incompatibility issues. Recommend testing via GitHub Actions CI which uses Xcode 15.x.
Does your PR add, remove, or rename any plugin parameters? No
Does your PR add or remove any graphical assets? No

Testing Notes

Windows (Validated ✅)

Build: Visual Studio 2022, x64 Release
Validation: pluginval v1.x, strictness level 5
Results: All tests passed (audio processing, automation, state save/restore, editor, bus configuration)

macOS (Limited Testing ⚠️)

Note: Local macOS testing was not possible due to environmental constraints:

Test environment: VMware Sonoma 14.7.2 + Xcode 16.2
Issue: NanoVG/Metal incompatibility causing image loading crashes (not related to this PR)
Official v0.7.13 release also crashes in this environment
Recommendation: Please test via GitHub Actions CI, which uses Xcode 15.x and should work correctly

The embedding code itself is platform-independent and uses standard C++ file I/O. The implementation should work on macOS, but comprehensive validation via CI is recommended.

Files Changed

NeuralAmpModeler.cpp: Embedding/extraction logic and helper methods (+260 lines)
NeuralAmpModeler.h: Added mNAMData and mIRData member variables (+8 lines)
Unserialization.cpp: Added embedded data extraction with versioned functions for 0.7.13 (+134 lines)

damiangr · 2025-12-13T21:06:53Z

macOS version was tested with Fix missing path separator in prepare_resources-mac.py and it seems that it works.

sdatkinson · 2025-12-14T01:12:44Z

Thanks for your patience with my review.

High-level question: what happens when the required files are gone (so the model is loaded from the serialized state), then you try to use the arrows? It feels like this is a new dimension that has to be handled. I don't really like the idea of being responsible for it in this repo.

A high-level criticism of the benefits of this: It seems to me that this gives half of what needs to be a whole solution to make sense--all of the cited benefits may be solved for the model files, but they remain unsolved for the main part of a DAW session--the audio (track) files? I don't really agree that a DAW session can truly be called "portable" if these two different file inputs aren't both handled.

I'll look at the lines and provide detailed feedback there. [There are a considerable number of changes in this PR.]

sdatkinson

My overall impression after reading the code is that this is too complicated for what it brings.

The changes below are probably good pieces of advice if you want to use them to improve your fork, but it's sort of surprising how many different places are touched in this PR. Unexpected things may be happening--for example, the sample rate validation.

If that's a problem in the existing code aside from the feature this is trying to introduce, then it'd be worth it to PR those fixes separately (i.e. validate inputs and demonstrate that those validations are failing). But if this is introducing invalid usage of those existing pieces of code, then it's not immediately clear to me how that's happening, and that feels risky for the stability of the plugin.

You definitely want to re-do the unserialization code to follow the existing pattern because that part definitely isn't following the pattern so far and risks breaking existing projects--which would be ironic, given the goal of this PR 🙂

.github/workflows/test-release.yml

.gitmodules

sdatkinson · 2025-12-14T01:19:37Z

NeuralAmpModeler/NeuralAmpModeler.cpp

    {
-      const auto irData = mStagedIR->GetData();
-      mStagedIR = std::make_unique<dsp::ImpulseResponse>(irData, sampleRate);
+      // IR is in error state - discard it


Is this wise? Feels wrong.

Removed unnecessary safety checks

NeuralAmpModeler/NeuralAmpModeler.cpp

sdatkinson · 2025-12-14T01:30:10Z

NeuralAmpModeler/NeuralAmpModeler.cpp

    std::unique_ptr<nam::DSP> model = nam::get_dsp(dspPath);
-    std::unique_ptr<ResamplingNAM> temp = std::make_unique<ResamplingNAM>(std::move(model), GetSampleRate());
-    temp->Reset(GetSampleRate(), GetBlockSize());
+    std::unique_ptr<ResamplingNAM> temp = std::make_unique<ResamplingNAM>(std::move(model), sampleRate);


Was there an issue here? Or is this function being called from a different state from before?

yes, there was an issue - sample rate could be 0 during unserialization causing division by zero.

Double checked and removed - reverted to use [GetSampleRate()] directly like upstream. No issue, this was an unnecessary change.

NeuralAmpModeler/NeuralAmpModeler.h

sdatkinson · 2025-12-14T01:32:41Z

NeuralAmpModeler/Unserialization.cpp

  mIRPath.Set(static_cast<std::string>(config["IRPath"]).c_str());

-  if (mNAMPath.GetLength())
+  // NEW: Prefer embedded data over file paths


I think I'd like the opposite, but I'm willing to hear your rationale for this.

Fixed. It will check for file first and revert to embedded data if not found.

NeuralAmpModeler/Unserialization.cpp

sdatkinson · 2025-12-14T01:37:08Z

NeuralAmpModeler/Unserialization.cpp


  mNAMPath.Set(static_cast<std::string>(config["NAMPath"]).c_str());
  mIRPath.Set(static_cast<std::string>(config["IRPath"]).c_str());



What's going on here is being done wrong. Since you're defining new logic on how to deserialize for the next version, you should be adding new functions for v0.7.14 similar to how the ones for 0.7.12 were added.

created proper _GetConfigFrom_0_7_13() and _UpdateConfigFrom_0_7_13().

damiangr · 2025-12-14T17:16:26Z

rows? It feel
Thank you for checking !
Regarding the arrows, they will now work as follows:
If the file is not found, but the directory exists, the arrows will work as usual to switch to the next file within the directory.
If the directory does not exist, the arrows do nothing, but the saved embedded files are loaded.

This enables DAW sessions to be fully portable by storing the actual NAM model and IR file data in the session, not just file paths. Changes: - SerializeState: Embed NAM/IR file data in session chunk - UnserializeState: Load from embedded data if file path not found - New _StageModelFromData/_StageIRFromData functions - Version bump to 0.7.13 with backward-compatible serialization Behavior: - Prefers loading from file path if available (for easy model updates) - Falls back to embedded data if file is missing (for portability) - Fully backward compatible with older session formats

damiangr · 2025-12-14T18:28:13Z

All review feedback addressed:

Removed test-release.yml and prepare_resources-mac.py changes
Removed unnecessary safety checks from _ResetModelAndIR
Reverted _StageModel to use GetSampleRate() directly
Removed mutable - files read at load time now
Created proper versioned functions (_GetConfigFrom_0_7_13, etc.)
Changed to prefer file path over embedded data
Using version 0.7.13

…- Store NAM model and IR file contents in plugin state for full preset recall - Add right-click context menu to save embedded data back to disk - Support loading embedded data from serialized state - Maintain backward compatibility with existing presets Changes in submodules: - AudioDSPTools: Add IR memory loading and WAV export - NeuralAmpModelerCore: Add string-based NAM model loading

damiangr force-pushed the feature/embed-clean-v2 branch from ba35a5e to 974ec7d Compare December 13, 2025 21:03

sdatkinson requested changes Dec 14, 2025

View reviewed changes

damiangr requested a review from sdatkinson December 14, 2025 14:28

damiangr marked this pull request as draft December 14, 2025 14:46

damiangr force-pushed the feature/embed-clean-v2 branch 2 times, most recently from cac9298 to 1233630 Compare December 14, 2025 16:47

damiangr force-pushed the feature/embed-clean-v2 branch from dc05b69 to 4996eaf Compare December 14, 2025 17:48

damiangr force-pushed the feature/embed-clean-v2 branch from 4996eaf to e3a830d Compare December 14, 2025 18:21

damiangr marked this pull request as ready for review December 14, 2025 18:27


		mNAMPath.Set(static_cast<std::string>(config["NAMPath"]).c_str());
		mIRPath.Set(static_cast<std::string>(config["IRPath"]).c_str());

Add support for embedding NAM models and IR files in session data #595

Are you sure you want to change the base?

Add support for embedding NAM models and IR files in session data #595

Uh oh!

Conversation

damiangr commented Nov 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Features

Benefits

Implementation

PR Checklist

Testing Notes

Windows (Validated ✅)

macOS (Limited Testing ⚠️)

Files Changed

Uh oh!

damiangr commented Dec 13, 2025

Uh oh!

sdatkinson commented Dec 14, 2025

Uh oh!

sdatkinson left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

damiangr commented Dec 14, 2025

Uh oh!

damiangr commented Dec 14, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

damiangr commented Nov 5, 2025 •

edited

Loading