We describe a real-time implementation of a multi-phase excited LPC speech coder using Bell Laboratories DSP chips. A major advantage of multi-pulse excitation is that it produces natural-sounding synthetic speech and avoids both pitch analysis and voiced-voiceless decision. Pulse amplitudes and locations are determined by a noniterative, analysis-by-synthesis procedure which minimizes a perceptual distance criterion. The multi-pulse analysis is performed in fixed-point arithmetic on two DSPs. The analysis yields four excitation pulses for every 5 ms of speech, which are necessary for producing high quality synthetic speech. A companion real-time LPC area-parameter analyzer using DSP chips was described previously by Hanson and Olive [J. Acoust. Soc. Am. Suppl. 1 69, S18 (1981)].