Exploring the Influence of Noise in Speech Emotion Recognition Devices for Internet of Thing

Abstract

With the development of the Energy Internet (EI), the application of smart grids has expanded from the industrial field to homes and individuals, which effectively promotes the development of home Internet of Things (IoT) devices. The research of the home IoT aims to improve the user experience, and the focus is on the intelligence of the device. The intelligence is inseparable from human-computer interaction (HCI). Speech emotion recognition(SER) uses machines to recognize emotions in human speech, which is an important part of HCI. However, noise usually greatly influences the recognition accuracy in HCI. In this paper, we conduct experiments on 16 types of common noise in the environment. We find that some types of noise influence the recognition effect significantly but some do not. We explain this difference from two perspectives—spectrogram and statistical characteristics so that we will gain more insight as to what type of noise will and/or will not influence the recognition accuracy of a particular SER task.