HMM and BPNN based speech recognition system for home service robot

This paper proposes a two-stage speech recognition system based on hidden Markov model (HMM) and back-propagation neural network (BPNN) for home service robot. Since a home service robot would interact with different users, a speaker independent and robust system should be developed. The recognition system we proposed contains two learning stages to build the models of words. In the first stage, the Gaussian mixture model (GMM) likelihood probabilities are calculated by HMM. And then, the probabilities are treated as the input units of neural network in the second stage. The home service robot, May-1 is designed and implemented for realizing the speech recognition system. The experimental results show that the robot can successfully complete follow-me, recognition of names, and recognition of rooms tasks in the RoboCup@ Home league competition.