Smart Home Notifications in Croatian Language: A Transformer-Based Approach

Abstract

Natural language generation is one of the essential components in achieving spoken interaction between a human user and a computer system. In this paper, we describe the process of generating notifications in Croatian language by employing an approach that relies on sentence transformer models. The solution described here extracts the key information from raw system events available in scope of the IoT platform and uses them to construct user-friendly sentences. The use-case scenario we present in this paper is related to a smart home environment. In this scenario, the generated notifications are used to construct spoken notifications using a speech synthesis component.