User-Defined Keyword Spotting Utilizing Speech Synthesis for Low-Resource Wearable Devices

Abstract

In this paper, we propose a novel keyword spotting (KWS) system for wearable devices that allows users to add user-defined keywords in quick and easy way. Adding keywords in KWS requires developing a new model to support them, where the model development takes a lot of work and time. To overcome this, we propose an approach that automates the entire development phase of a KWS model for low-resource devices. The proposed system is characterized by automating the data collection step and training step using synthetic speech data. Our implementation and experiments show that the proposed system can automatically develop a user-defined KWS model within a minute.