Low-rate speech coding based on time-frequency interpolation