To actuate an in-vehicle device by speech without applying too much processing load to the in-vehicle device.

In the speech actuation system, a speech recognition section of a terminal device recognizes speech, a recognition result transmission section of the terminal device transmits the recognition result to the in-vehicle device, and an actuation control section of the in-vehicle device controls actuation of the own device based on the recognition result transmitted from the terminal device. When a specified user uses a mobile terminal device, a learning processing section of the mobile terminal device learns a speech characteristic of the specified user, and a speech extraction processing section extracts speech of the specified user from input speech based on the learning result.