It internally decodes as float and then converts to s16 by a call to
float_to_int16(). The caller can do this just as well by using lavr.