Speech-Recognition ICs Serve Consumer Electronics Needs

Two speech-recognition chips are available to serve the needs of cost-sensitive consumer electronics, as well as higher-end consumer products. The RSC-264T (see block diagram) is designed to minimize end-product cost while offering improved recognition accuracy. At a price of less than $2.25 each in OEM quantities, the chip is said to enable sales of interactive speech products for as low as $15. Meanwhile, the RSC-364 IC offers high-accuracy speech recognition with an error rate that’s half of the firm’s earlier offerings.At the heart of the RSC-264T is a specially designed 8-bit MCU that provides programming flexibility while supporting a complete suite of speech and audio technologies. These include speech recognition, speech and music synthesis, speaker verification, and audio recording and playback. The IC offers a full complement of features, such as a microphone preamplifier, that work to reduce system cost compared to the firm’s earlier products. A sophisticated on-chip neural network learns to classify sound data. The IC’s speech-recognition algorithms reach an accuracy of greater than 96% for speaker-independent recognition and greater than 99% for speaker-dependent recognition. The neural-network approach is said to eliminate the need for costly signal processing or extensive RAM. Complete systems may be built with just a few additional parts.In the RSC-364, a 4-MIPS MCU is teamed with an audio preamp, ADCs and DACs, a watchdog timer, 64 kbytes of ROM, and 2.5 kbytes of RAM. It sells for less than $5 each in OEM lots. Both ICs are supported by a comprehensive suite of development tools.