Continuous speech recognizer for low-end embedded devices

This paper discusses the implementation of continuous speech recognition system for low-cost bare-metal platform. The applied algorithms are well known, however they are optimized for constrained embedded environment. System is very flexible since it uses publicly available acoustic and language models, and dictionaries, and as a result of recognition produces readable ASCII string through serial port. Performance measurements show that it is slower compared to the same algorithm running on desktop, but it keeps its recognition accuracy.