Standard Object Systems, Inc. (SOS) will use its existing phonetic speech recognition (SPRS) tool kit to develop an advanced audio interface for speech recognition in a high noise military environment. A requirement exists for audio command, control, and communications systems in a noisy environment based on natural interfaces, such as speech recognition, which require no training for efficient intuitive operation. That existing speech recognition systems fail in even moderate noise environments was identified as a critical issue at the military workshop on Spoken Human-Machine Dialogue in 1995. As a potential SOS product, noisy speech recognition is of great commercial value in vehicles, offices and factories. To reduce noise effects in speech recognition, SOS will modify the SPSR processing in three ways: Add an adaptive filter to remove noise prior to speech processing; compute auditory based EIH data to create robust speech recognition features; and apply a genetic optimization algorithm imitating the human evolution of speech to improve noise rejection. These changes will be tested, with both off the shelf commercial speech software and the SPSR system, in a digital audio laboratory with controlled speech and noise to evaluate affects on speech recognition performance
