Navy Air Traffic Control (ATC) trainers use simulations of airspace traffic to improve skills and to provide realistic, safe training environments. Real-time continuous speech recognition is used to interface directly with ATC training devices, thus eliminating the need for human pseudo-pilots or sim-pilots. With the expertise gained in constructing ATC training devices utilizing speech recognition technology, the Navy has applied speech recognition to a real time Navy Carrier Air Traffic Control (CATC) application called ISIS (Integrated Shipboard Information System). ISIS is located in three CATC areas, Air Operations, Primary Flight Control, and Carrier Controlled Approach (CCA), all with varying noise levels. Operationally, the ISIS speech recognition system must maintain near perfect recognition accuracy while being operated in a high noise environment.

In general, speech recognition syntax is constructed with a balance between the phraseology and discernment between similar sounding words. Furthermore, syntax development for a particular users/group phraseology demands an intimate understanding of the users/group functional requirements. Beyond what is required for basic recognition, the ISIS recognizer uses an additional noise rejection algorithm which models one's speech, compares the speech input with the model, and competes the result with real-time templates whereby sounds other than the intended speech can be detected. Using this noise rejection model in a different manner, the ISIS speech recognizer provides an out-of-phraseology capability that filters incorrect words and phrases. Exhaustive testing of a particular syntax with and without a noisy background is necessary to establish a desirable noise rejection threshold. Finally, this report discusses the hardware, software, syntax development, noise studies, out of phraseology algorithm, results of the fielded system, and implementation of the ISIS speaker dependent speech recognition system.