This paper presents the M-LCELP (Multi-mode Learned Code Excited LPC) speech coder, which has been developed for submission to the competition for setting the North American standards for half-rate digital cellular systems. M-LCELP develops several techniques to achieve high-quality synthetic speech. The MOS subjective test shows that 4.075kbps M-LCELP synthetic speech quality is high, equivalent to that of an 8kbps full-rate VSELP coder.