A VLSI wordprocessing subsystem for a real time large vocabulary continuous speech recognition system

The authors present the architecture of a word-processing subsystem for a large-vocabulary real-time continuous-speech-recognition system. The system contains three application-specific integrated circuits that perform the Viterbi algorithm for 50000 states using a breadth-first recognition search in real time. This corresponds to a computation rate of 225 million operations per second, excluding memory operations where 670 Mb have to be accessed per second