A Fast-Match Approach for Spoken Language Systems in Medical Environments

Robust speech recognition is needed for speech technology to be useful in the real world. People's speech is often spoken in noisy environments which greatly increases the difficulty of hihg-accuracy speech recognition. The research proposed here will result in the identification of technologies for speech recognition in noisy environments, as well as a prototype system designed to automatically segment speech in noisy environments. Phase I has two primary goals: first, to identify which of a number of signal enhancement algorithms and signal processing techniques are best-suited for the task of segmenting noisy speech; and second, to use these technologies in a prototype system which robustly discriminates between speech and non-speech. Artificial Neural Networks will provide the core discrimination engine in the prototype. If successful, the technology proposed here will be used as a pre=processor to a recognition system which recognizes medical phrases and user interface commands. Such a system can be integrated with numerous applications to allow hands-free data entry and command and control in a wide variety of noisy environments. Anticipated Benefits: Spoken language systems will become an integral part of our lies in the near future. Using speech as an interface will allow hands-free control of computer systems.