1. Although the prototype is built and there is not enough time digging the Neural Net in Arduino. It is still worth introducing Neurona library as powerful tools in voice reorganization in Edge Computing.

Import Neurona from library manager .

There is one example on color Senor.

#include <Neurona.h>
#include <LiquidCrystal.h>
MLP mlp(NET_INPUTS,NET_OUTPUTS,layerSizes,MLP::LOGISTIC,initW,true);

 

2. In Multi-Layer Perceptron - an implementation in C language, the neural network learning is introduced. Tool used for training is provided to.

3. How to use.

It can be used from above UI, and preset the layers. Or program in github can be downloaded and run locally. Simple as it is, the final goal is to train customized data array for Weight (W) .

In this example, the initial weight is given.

double PROGMEM const initW[] = {2.753086,-11.472257,-3.311738,16.481226,19.507006,20.831778,7.113330,-6.423491,1.907215,6.495393,-27.712126,26.228203,-0.206367,-5.724560,-22.278070,30.065610,6.139262,-10.814282,28.513130,-9.784946,6.467021,0.055005,3.730361,4.145092,2.479019,0.013003,-3.582416,-16.364391,14.133357,-5.089288,1.637492,5.894826,1.415764,-3.315533,14.814289,-20.906571,-1.568656,1.917658,4.910184,4.039419,-10.848469,-5.641680,-4.132432,10.711442,3.759935,19.507702,17.728724,-3.210244,-2.476992,8.988450,5.196827,2.636043,17.357207,2.005429,11.713386,-5.453253,-6.940325,10.752005,0.666605,-7.266082,-3.587120,-9.921817,-12.682059,-15.456143,-13.740927,0.508265,15.179410,-11.143178,-19.085120,1.251235,22.006491,-4.227328,-0.444516,3.589025,0.649661,13.675598,-13.026884,-11.229070,-15.300703,-1.718191,6.737973,-28.176802,-2.505471,5.197970,7.007983,-2.869269,3.650349,18.029204,4.098356,10.481188,-2.566311,9.927770,2.344936,4.524327};

You can follow the instruction to make your training model for your calculation.

 

In the Blowing-Whistle as Controller Project, more data shall be needed to trained for useful value of initW[], the output can be used for authentication Purpose.

To find out which Whistle is right One before assign value True to variable  Auth_Pass,

In this way, the deep learning model can be deployed in arduino board instead of unstable Over-the-Air Cloud End API.

It has been proved by MKR1000 is qualified for Edge Computing. Hopefully, I can verify the validity of the application of  Machine Leaning on Blowing-Whistle as Controller,then.