A speaker includes a diaphragm having a principal vibration part and annular internal and external supports for supporting the principal vibration part, a large-diameter voice coil unit mounted at the outer edge of the principal vibration part, a small-diameter voice coil unit mounted at the inner edge of the principal vibration part, and a yoke integrated with the diaphragm. Positioning members fixed to the diaphragm as well as secured to a predetermined site of the yoke are provided for restricting the positions of the diaphragm relative to the yoke in a radial direction and an axial direction perpendicular to the radial direction so that the large-diameter voice coil unit and the small-diameter voice coil unit are arranged within respective magnetic gaps in a magnetic circuit.