Google has launched a web-based tool named, ‘Teachable Machine’ , which lets you do deep learning tasks, without having to write any piece of code. The tool allows you to teach the machine, a simple pattern recognition.
Using Teachable Machine, you can perform 3 types of tasks:
1.Classification based on image
2.Classification based on pose
3.Classification based on audio
Image Classification is a task of assigning labels to image. The image is considered as a whole for the classification. The system learns patterns from colors to classify different images.
Audio Classification is a task of assigning labels to different audios. The system generates audio image from audio frequencies and uses those images to learn audio patterns.
In pose classification, the system locates human from the whole image and then generates shape-based features of a particular human pose. The system then uses this pose features, in order to classify.
If you want your machine to recognize certain actions, you would have to train the machine. It’s almost similar to teach an infant, to recognize the voice or face of a parent.
The process is achieved in 4 steps:
2.Train the system to recognize pattern from gathered dataset
3.Testing and tweaking the trained system
4.Integration with an application
First, we need to gather or generate certain samples, from which the machine can learn patterns. For example, I am going to make a smart Tic-tac-toe game, that can be played through voice commands. The system will recognize the position of the sign where we want to mark it.
In tic-tac-toe, there are total 9 positions which can be marked. Accordingly, we need to record audio samples for those 9 positions. We will also have to record another type of audio, which would act as the Background Noise. So that, when nothing is spoken, it would be considered as Background Noise by the machine.
Let’s gather some samples.
Go to Teachable Machine and select “Audio Project”.
First, record some background noise.
After you are done with that, record numbers 1,2,3,4,5,6,7,8,9 in different class sections. Remember to give as many sound samples as possible, so that the system can understand numbers properly.
Note that the system will recognize the audio from which it has been recorded. If you want to recognize the audio from different devices, then add sample audios from those devices too.
Teach the system to recognize patterns
After gathering the audio samples of each number, we teach the system to recognize those audio samples. This process of learning is called Training. In this training process, the system recognizes complex patterns from the samples and tries to memorize them.
Click on the “Train Model” button to start training.