//forked from Andreas Refsgaard

let video;

let label = "waiting...";

let displayLabel = "waiting...";

let previousLabel = "waiting...";

let confidence = 0.0;

let classifier;

let modelURL = 'https://teachablemachine.withgoogle.com/models/9ZhtUP8P/';

let handSound;

let personSound;

​

// STEP 1: Load the model + sounds 

function preload() {

  classifier = ml5.imageClassifier(modelURL + 'model.json');

  handSound = loadSound("shock.mp3");

  personSound = loadSound("suspense.mp3");

}

​

function setup() {

  createCanvas(640, 520);

  video = createCapture(VIDEO);

  video.hide();

  classifyVideo();

}

​

function draw() {

  background(0);

  image(video, 0, 0);

​

  // STEP 4: Playback sound if there is a switch in labels

  if (label == "hand" && previousLabel != "hand") {

    handSound.play();

    personSound.stop();

  } else if (label == "person" && previousLabel != "person") {

    personSound.play();

    handSound.stop();

  } else if (label == "nothing") {

    //don't do anything on this label

  }

  //Display current label

  textSize(32);

  textAlign(CENTER, CENTER);

  fill(255);

  text(displayLabel + " " + confidence, width / 2, height - 16);

  //Update previousLabel

  previousLabel = label;

}

​

// STEP 2: Do the classifying

function classifyVideo() {

  classifier.classify(video, gotResults);

}

​

// STEP 3: Get the classification

function gotResults(error, results) {

  if (error) {

    console.error(error);

    return;

  }

  // Store the label and classify again

  confidence = nf(results[0].confidence, 0, 2);

  //Small hack: Only update labels when confidence is high

  if (confidence > 0.95) {

    label = results[0].label;

    displayLabel = label;

  } else {

    displayLabel = "unsure";

  }

  classifyVideo();

}

​