The EarAl software is designed to help people who are suffering from hearing disability or even people that would like to use headphones but depends on the circumstances which they are in, such as the need to be aware of the things that are going around in their surroundings, cannot wear them. To do so, the software consists of five different major parts, the UI, the main core, the input module, the filtration unit, and the notification module.
The UI of the software is designed as easy as possible to use, there are predicted values of the YAMNET pre-trained network and also the values which are being predicted with the software and are based on those predicted values of the YAMNET pretrained network, these values will be showing very clearly on the screen. There is an option, a microphone choosing menu, for those who’d like to connect their Bluetooth headset microphone and place it somewhere else and still wish to use their phone. Besides, there is a menu to choose which notifications the user wants to be notified with.
The main core is based on YAMNET pre-trained network. It classifies the input into 521 different classes as its output. The input module is a converter that changes the input to be the same size and specification as the YAMNET model input.
In the Filtration unit a matrix is filled with the different predicted values that the main core provides, then, the software prediction value is guessed with this matrix.
And last but not the least, the notification module notifies the user when the chosen situations happen.
The app is made in the unity 3D engine and Barracuda is used as the library to read the ONNX model.