To improve the awakening effect of an occupant by detecting a dozing state of the occupant according to a state of voice detected by collecting/ analyzing a car inside sound, and outputting this as voice by generating a dialogic sentence on the basis of voice information on the occupent according to the dozing state.

A car inside sound is collected by the car inside sound collecting part 1 to detect whether or not voice is contained by analyzing information on a car inside sound collected by the voice analyzing part 2, and language is extracted so that this is recorded in the language recording part 3. An occupent state is detected by the occupent state detecting part 4 on the basis of a voice detecting signal. A prescribed dialogic sentence is successively generated by the dialogic sentence generating part 6 by using the language recorded in the language recording part 3 according to the occupent state so that this dialogic sentence is outputted from the voice output part 7. According to this constitution since a dialogue with an occupent is repeatedly taken while successively changing the dialogic sentence according to a degree of a dozing state of the occupent, the occupant does not get used to the voice so that an awakening effect can be sufficiently exhibited.