TERI RESEARCH INC  Providing the Information Advantage...                home.gif (2807 bytes)     

SERVICES       DEFENSE APPLICATIONS      KIOSK-COMMERCIAL       AMERICAN DISABILITIES ACT      back.gif (114 bytes)

Continuous Speech/Fixed Grammar

Grammar and context algorithms when incorporated in the continuous speech grammar recognizer provide Signal-to-Noise Ration (SNR) processing gain in high-noise environments for continuous speech recognizers. The reason speech grammar recognizers perform with higher accuracy, compared to just continuous speech models, is because the grammar and context algorithm feedback from knowledge based databases supports interpretive understanding.

The fundamental "building blocks" of Natural Language Speech (NLS) are contained in the 60,000-word vocabulary, syntax and semantics of the language corpus. An Automatic Speech Recognizer (ASR) requires "apriori" knowledge of these language building blocks, to perform its recognition function. Additionally, the ASR is initially programmed to anticipate a subset of the language building blocks (albeit large) to process the information in real-time and with sufficient accuracy. The speaker (user) and the listener (ASR) must have the same vocabulary and semantic knowledge for any communication. However, a mutually negotiated syntax (grammar) helps match or compensates for the disparity in the user application familiarity and language training. Continuous speech ASRs are used and structured to recognize either constrained or unconstrained syntax. These syntax combinations or word sequences, can be defined as both domain relevant sequences and natural language grammar. In other words, the ASRs can be programmed to recognize variable syntax that contain both grammatically correct and domain unique constructs. The Grammar Model Database contains all of the selected Domain and Rhetoric permitted syntax together with the inherent ASR natural language grammar. This approach lends itself to a grammar-based syntax that is universally accepted by both dictation and command type speech recognition.