Blog Post

Internet of Things Blog
13 MIN READ

Azure Percept Audio – Custom Keywords and Commands

Peter Gallagher's avatar
Jul 07, 2021

In the latest in my series of blog posts on the Azure Percept Developer Kit and specifically around the Audio System on a Module  (SoM), we look at how we can create and use Custom Keywords and Commands with the Audio SoM.

 

Previously

 

In the previous post - Azure Percept Audio - First Steps - We got started with the Azure Percept Audio System on a Module (SoM).

 

We deployed a sample Hospitality application and saw how we could issue commands to the Percept DK, how it would respond, and how the Sample Application would simulate the effects of the commands on screen.

 

 

In this post we’ll take the sample further by training our own Custom keyword and creating a Custom Command.

 

Custom Keywords

 

For the Azure Percept Audio, a Keyword is the word that the Percept DK“listens” for in order to begin listening for commands. Sometimes referred to as a “Wake-Word”, this word defaults to “Computer” in the Hospitality Sample.

 

We have access to some pre-trained keywords deployed along with the sample, allowing us to choose from;

 

  • Assistant
  • Abigail
  • Computer
  • Jayden

Which we can set by pressing the “change” link next to the “Custom Keyword” item below the toolbar area;

 

 

Hitting the Custom Keyword Change Link shows us the “Change custom keyword” flyout with the various options available to choose from.

 

 

Selecting another custom keyword and pressing the “Save” button at the bottom of the dialog will update the Percept DK, and we can then wake the device with the new Keyword.

 

Interestingly, I found that using the alternative keywords of Assistant and Abigail, allowed the Percept DK to wake faster than with the default of "Computer", I'm not sure why this is to be fair, but it's possibly due to the number of syllables in the word maybe?

 

Training our own Custom Keywords

 

In case none of the built in Wake Words are suitable, we can also train our own custom keywords to wake the device. We can do this directly from the Sample Application, allowing us to use a word of our own choosing;

 

 

Pressing the “Create Custom Keyword” button in the toolbar at the top of the page allows us to configure and train our own custom keyword;

 

Here we can enter the keyword of our choosing, select an Azure Speech Resource and a Language, and pressing save will begin the training process.

 

 

Bear in mind however, that the keyword must have between 2 and 40 syllables to be accepted;

Updated Jul 07, 2021
Version 2.0
No CommentsBe the first to comment