A Real Time Connected Speech Recognition System Using Hidden Markov Models

This paper describes a PC-based system for recognition grammar-constrained connected speech in real time, using Hidden Markov Models. The heart of the system is the BT-100 parallel processor that can grow in computation power as the difficulty of the speech recognition task increases in terms of vocabulary size, language complexity and speaker independence. Real time speech recognition has been achieved with this system for a 70-word data entry task and a 132-word airline reservation system. Recognition results obtained are also presented.