Mature Speech Recognition Technology for Military Helicopters

This paper summarizes the history and development of UK speech recognition technology for helicopter applications sponsored by the UK Government. The problems associated with designing successful automatic speech recognition systems (also referred to as Direct Voice Input systems) are discussed, including the early use of speaker-dependent whole-word systems in Wessex and Lynx helicopters that required a lot of investment of each speaker’s time to train personal speech models. More recent results are then covered for the use of a robust and accurate phonetically-based speech recognizer Aurix® asr in a ground-based simulator with full vibration and noise, where a word accuracy of 95% was obtained. Aurix asr proved much more robust to changes in the speaker’s voice, as well as eliminating the need for a large investment in training. Aurix asr has also been used in the US Army’s Virtual Cockpit Optimization Program simulator through a US/UK government bi-lateral agreement for the exchange of helicopter research information. In this application DVI has been integrated into a model of a UH - 60 Blackhawk Helicopter in the Battlefield Highly Immersive Virtual Environment dome at the AMRDEC Advanced Prototyping, Engineering and Experimentation Laboratory.