The embodiment of the invention provides a speech interaction method, a speech interaction device and a robot. The method is applied to the robot, and comprises the steps of obtaining an image when the sound source angle of a received sound signal is in a preset angle range of the robot, and recognizing angles of one or more human faces in the image; selecting the person whose human face angle is the closest to the sound source angle as the speaking person; adjusting the angle of the robot to make the center of the face of the speaking person falls in the center of the front of the robot, so that the sound signal is responded to. According to the speech interaction method, the speech interaction device and the robot, the speech interaction function of the robot can be more intellectualized and personified.