If you didn’t know before, one of Edge Impulse’s many deployment features includes deploying your trained impulse straight from the Studio to the web browser on your phone — without having to write any code! This enables you to quickly create a prototype application using the Edge Impulse WebAssembly library generated from your trained project, and also allows you to verify that your keyword spotting or audio classification model works as intended in the field, on a real edge device, in real-time.
When you are classifying audio — for example to detect keywords — you want to make sure that every piece of information is both captured and analyzed, to avoid missing events. This means that your device needs to capture audio samples and analyze them at the same time. Edge Impulse’s web-based continuous audio classification application utilizes a moving average filter to smooth out your incoming audio data and remove random noise, the filtered audio signal is then passed into the inference classifier function of your model’s Edge Impulse library to get the resulting prediction of what your phone just heard (for example, the keyword “Hello World” or just ambient environment noise).
How do I get started?
To start, create and train a model to classify human or non-human audio by following one of our tutorials:
Then, to deploy your audio classification model to the edge on the web browser of your smartphone, select the “Devices” tab of your Edge Impulse project:
Click on the Connect a new device button and then select Show QR code next to “Use your mobile phone”:
Scan the generated QR code from the camera application on your smartphone:
On the bottom of the page on your phone, click on the Switch to classification mode button to start downloading and building your proof of concept continuous audio classifier application using the Edge Impulse SDK directly on the web. When the project is finished building, click Give access to the microphone to allow the web browser to use your phone’s microphone:
Now see your web-based continuous audio classifier in action!
