Interpreting voice commands in radar systems Recognizing speech of multiple languages with limited resources in an offline environment

Abstract

This master’s thesis is done in collaboration with SAAB and outlines an architecture for developing a voice command system with focus on supporting multiple languages, ease of introducing new domain knowledge, resource efficient and running in an offline environment. The solution and the decisions made along the process as well as their advantages and drawbacks are explained in this report. As basis for the project we have worked with CMU Sphinx and Grammatical Framework to aid us in the process. Furthermore, the US-English community language model available from CMU Sphinx has been adapted to better understand the Swedish accent with the help of a tool provided by CMU Sphinx. The result of the adaptation as well as recommendation for future development is also included in this thesis.