This paper describes a hands-free speech/sound recognition system developed and evaluated in the framework of the CompanionAble European Project. The system is intended to work continuously on a distant wireless microphone and detect not only vocal commands but also everyday life sounds. The proposed architecture and the description of each module are outlined. In order to have good recognition and rejection rates, some constraints were defined for the user and the vocabulary was limited. First results are presented; currently project trials are underway.