Teachable Machine

Teachable Machine

The buzzword machine learning depicts mystery yet intrigue to most. To those who dare to venture into the unknown, it takes weeks or even months to train an accurate model with large amounts of data to solve the "simplest" question (e.g., what a cat looks like?). However, with Google's Teachable Machine, machine learning--unsupervised, supervised, and reinforced--simplifies and trains data in a matter of seconds!

Overview

In short, machine learning (ML) is an approach to teach computers to make decisions and predictions based on data. Teachable Machine (TM) is a web-based tool that introduces beginners to the fundamentals of ML, a field of artificial intelligence (AI).

Rather than thousands of data around the world that may require annotation, the data comes from you and your browser. In this article, we will step through training a model to detect and classify sound.


Part 1: Training

  1. Visit https://teachablemachine.withgoogle.com/ and click Get Started in the upper right corner.
  2. Select Audio Project as the model to train.

    Screen-Shot-2021-06-26-at-2.39.05-PM
  3. To teach it, you need to create at least 3 classes. Rename the classes to the sound you are recording. In this example, they are humming, keyboard typing, and mouse clicking.
  4. Record some samples by clicking on Record 2 Seconds, Recording, and then Extract Sample.

Screen-Shot-2021-06-26-at-2.46.45-PM

  1. Once you have sufficient data samples, hit Train Model on the right.
  2. Test it! Try a variety (e.g., fast and slow keyboard typing). ML has edge cases, and it's up to you to find and fix them.

Screen-Shot-2021-06-26-at-2.52.19-PM

If your model has bugs, add more training data with augmentation in mind. Augmentation dramatically increases a model's accuracy and decreases overfitting. Remember, your model is only as good as your data set.


Part 2: Exporting

TM hosts your model on their servers, so you can use it whenever you'd like! Export the model by copying the code to your Qoom project's HTML file between the <body> tags.

Screen-Shot-2021-06-26-at-2.58.37-PM


qoom

You have trained and developed a ML model with zero code in your browser! Further your model by adding more classes, and go crazy with training data.

We cannot wait to see what you created with Google's Teachable Machine. Make sure to share in Qoom's Discord server and showcase!


Check it out here: https://punycat6.qoom.space/~/spK !