Design of a Speech Anger Recognition System on Arduino Nano 33 BLE Sense

Speech Emotion Recognition (SER) has gained growing popularity due to its wide applications in almost every available field. Past work has been done on large-scale processing boards with a variety of extracted features. Many different methods have been proposed for the SER system but have lacked aspects in either size, complexity, or recognition accuracy. Due to limitations on size and resources, designing a system that can overcome these drawbacks becomes imperative. The solution is to design a system that is small, accurate, and economical. Using Tiny Machine Learning and SER is the best solution since it can be done on a small-scale and relatively high emotion recognition rate. This paper presents past work, the hardware, software, and the SER prototype’s field design, focusing on detecting the variations of the Anger emotion. A simple and optimum CNN architecture was developed for Arduino Nano 33 BLE Sense implementation. Prototype validation showed that our system could detect not angry, about to be angry, and angry emotions.