Intent Classification

There are often a lot of ways to say the same thing: “boot up”, “power on”, and “fire up” are all different ways that someone might say “turn on”. To handle these cases, we can use an IntentClassifier. This type of voice controller allows you to define actions in natural language that you want users to be able to invoke with a variety of different possible commands.

When the user gives a command, the IntentClassifier tries to invoke the most semantically similar action that you specify:

// command handlers map voice commands to the actions they should trigger
const commandHandlers = {
    "turn on": () => { // this function triggers when the user says something similar to "turn on"...
        document.querySelector("body").style.backgroundColor = "white"
    },
    "turn off": () => { // ... and this one triggers when they say something similar to "turn off"
        document.querySelector("body").style.backgroundColor = "black"
    }
}

const intentClassifier = Moonshine.IntentClassifier(commandHandlers)

var transcriber = new Moonshine.MicrophoneTranscriber(
    "model/tiny",
    intentClassifier,
    true // VAD enabled
)

transcriber.start();

Now when the user says something similar to “turn on” or “turn off”, the most relevant action will be triggered.

Note that we enabled voice activity detection—this is highly recommended when you are using voice control.