Generate Audio Plugin with Neural Network

Is it possible to create audio plugin that uses neural network?

We trained a DNN network and saved it to matlab file.

In Audio Plugin class, inside

function setupImpl(plugin, u)  

we tried to load our pretrained network by using

load('trainedNetwork.mat');

but "The function 'load' is not supported for standalone code generation. Use coder.load"

So we tried using

coder.load('trainedNetwork.mat');

but "Found unsupported class for variable using function 'coder.load'. The value at 'matlabObj.net' is unsupported for code generation".

Our DNN network consists of:

layers = [ ...
    SequenceInputLayer;    
    fullyConnectedLayer
    reluLayer
    fullyConnectedLayer
    reluLayer
    fullyConnectedLayer
    tanhLayer
    outputLayer;
];

where SequenceInputLayer appears to be not supported for code generation.

Is is possible to generate VST that uses neural network?

[Max Henry]

I'm in the same boat -- I made it as far as using:

coder.loadDeepLearningNetwork('trainedNet.mat');

but I'm getting the same issue: the value at 'matlabObj.net' is unsupported for code generation".

[Adrian Klimczak]

That error message probably occurs due to neural network containing layers that are not supported for code generation. We were not able to reach satisfactory results with Matlab Neural Network implementation for audio, therefore we decided to use TensorFlow.