Advanced Audio Interfaces

We will demonstrate an advanced audio interface for reliable automatic speech recognition (ASR) and other speech systems in dynamic, noisy environments. Our approach uses an innovative, high performance, multi-stage noise removal front-end, that can be integrated with a variety of final speech applications. We will demonstrate end-to-end performance from integrating this processing with multiple commercial-off-the-shelf (COTS) ASR systems. Our noise removal processing will include adaptive noise cancellation, adaptive noise power estimation coupled with nonlinear noise prefiltering, cancellation, adaptive noise power estimation coupled with nonlinear noise prefiltering, and a high performance artificial neural network (ANN) noise removal post-filter. Equalization of channel transfer variations will also be included. Our ANN processing has already been shown to give much greater gains in ASR noise resistance than any other approach to date. From prior results, we can expect greater than a 30 dB gain in ASR resistance to background noise, even in highly dynamic noise environments. Performance will be demonstrated on aircraft maintenance and repair ASR applications in dynamic, high-level, flight line noise environments. Phase I will develop and demonstrate the basic methods and their end-to-end ASR performance. In Phase II a complete real-time system, with documentation and training, will be delivered for USAF testing and retention.