Human-robot-interaction using cloud-based speech recognition systems
Christian Deuerlein, Moritz Langer, Julian Seßner, Peter Heß, Jörg Franke
Abstract
Progress in natural speech processing has enabled significantly more powerful speech processing systems, primarily due to the use of machine learning technologies. In order to integrate cloud-based speech recognition systems for human-robot interaction, an interface for the voice control of a lightweight robot was developed. The main contribution of this work is the design and implementation of a software interface to recognize commands via cloud-based speech processing and the subsequently conversion into machine-readable code. Requirements for the evaluation of different cloud-services for the control of robots are determined. Furthermore, the control architecture for the robot is modeled and implemented. An example application, which enables users to control robot movements via speech, is realized as a proof of concept and for additional studies. This application includes the basic features of cloud-based speech processing: intent recognition from utterances, slot filling and dialogue-based interaction. Lastly, the influence of background noise on process safety was examined within an experiment. It turns out that a feasible process reliability can be achieved with the system despite the presence of background noises.