Noise-Robust Speech Recognition to Support Maintenance Activities

Recent advances in computer recognition of human speech have made it possible to perform reliable recognition of the content (words) in continuous speech. We propose to use these technological advancements, together with the expertise we have gained from developing a number of reliable, easy to use DOD systems that automatically respond to natural, spoken human commands. These systems reduce required human effort, improve performance, and increase the timeliness of information delivery. Specifically, we will adapt to the aircraft maintenance application several of our existing and emerging systems, including our system for speech recognition in noisy environments currently under development for the U.S. Army. In particular, we will apply a combination of proven and novel analysis techniques to filter out band-limited, broad-spectrum, and impulse noise. Our objectives will be to:. Design a flexible, highly accurate speech recognition system that is usable in high noise environments to support aircraft maintenance repair activities. Adapt and implement noise reduction algorithms resulting in relatively noise-immune speech recognition. Evaluate these noise reduction algorithms and the resulting speech recognition performance; and . Demonstrate an early prototype system.