Neural-network speech processing for toys and consumer electronics

The ongoing challenge in speech research is the recognition of continuous, unconstrained speech. In comparison, isolated word recognition with small vocabularies is easy. Many research efforts and commercial ventures aim at the high end problem. Sensory Inc. has successfully focused on the low end, producing a family of inexpensive speech recognition ICs for markets such as interactive toys, consumer electronics, electronic learning aids, telecommunication devices, security systems, and household appliances. These markets are extremely cost conscious and demand solutions in the under five dollar price range. Previous low cost solutions relied on coarse properties of the speech signal, such as the amplitude envelope or zero crossing rates, and typically yielded poor recognition performance. Digital signal processor based approaches can achieve adequate performance but are beyond the low cost market's reach. Sensory uses a neural network based approach that provides robust performance at a low cost. Sensory's general purpose RSC-164 controller performs four speech recognition functions.