Emovo: A real-time anger detector on the smartphone using acoustic signal

Abstract

Anger detection plays a pivotal role in mitigating potential mental health risks, presenting a significant yet challenging task. An anger detection app helps better track users' emotional scenarios and helps to prevent potential mental illness. Such system on smartphones requires less computational resources, but many lite emotion recognition systems are based on facial expression as input, while facial data is difficult to obtain in the daily use of smartphones. We propose Emovo, a real-time anger detector on the smartphone using the acoustic voice signal. Emovo uses audio as input, which makes it easier to collect audio data through microphones in daily usage and easier to implement with fewer parameters. Our model on smartphones has a comparable accuracy of around 84.3% on the public dataset and has a number of parameters of around 0.2M. We explore the effects of variables such as gender and semantics through user research. Emovo stands as a compelling alternative for effective emotion recognition on smartphones.