As a speaker produces a sound, the positions and angles of specular reflection plates put on the skin of the vocal organs and periphery thereof change. Resultantly, a ray of light applied from a light source section is reflected by the specular reflection plates uniquely in accordance with a law of reflection and the direction of the ray of light changes. Specularly reflected light spots moving on the face of a position detection sensor are detected by the position detection sensor, thereby detecting the positions of the specularly reflected light spots corresponding to the shape of the vocal organs and periphery thereof produced as the speaker produces the sound. Time characteristic and position change features are extracted based on the specularly reflected light spot positions detected. The features are compared with standard features patterns, thereby classifying input voice.