Using multiple HMM recognizers and the maximum accuracy method to improve voice-controlled robots

This paper proposes a speech recognition method using multiple speech recognizers with a combination of techniques to enhance robot control. The procedure consisted of 2 parts: 1) recognizing robot commands using multiple Hidden Markov model (HMM) recognizers, and 2) sending recognized commands to control the robot. In the first part, which is the main focus, three speech recognizers using gender-independent, gender-dependent and gender HMMs were applied to identify robot commands from speech signals. Then, the speech recognition results were combined using the maximum accuracy method. In the second part, the combined speech recognition results were used to control the robot with computer-robot control hardware. When using single-syllable robot commands, the results showed that without the combination method, average robot command recognition rates of 99.0%, 99.2% and 98.8% were achieved using gender-independent, gender-dependent and gender acoustic models, respectively. After the combination method was applied, the results revealed that the gender-independent and gender acoustic model speech recognizers, with the maximum accuracy method, could obtain a high robot command accuracy level of 99.4%, when the gender of the user was not known in advance. When the gender of the user was known beforehand, a robot command accuracy of 100.0% (no error reported) was achieved using three speech recognizers with the maximum accuracy combination method. When using single- and bi-syllable robot commands, high average recognition rates could also be achieved. The results indicate that robot command recognition using the maximum accuracy method with three kinds of HMM speech recognizers is an efficient method to enhance robot control by human voice.