A voice connection system comprises a sensor matrix sending to a control unit signals indicating an animal's status in terms of stimuli, feelings, events, actions or behavior of said animal. The control unit selects--depending on the signals received--a voice message recorded into memory means associated to the unit, and activates a loudspeaker belonging to the system, in order to issue said voice message. The system further comprises voice recognition means, which activate pulse generators depending on the content of a voice message uttered by the human user.