Coding of voiced speech by extraction of the glottal waveform has shown promise in improving the efficiency of speech coding systems. This thesis describes an investigation into the performance of such a system. The effect of reverberation on the radiation impedance at the lips is shown to be negligible under normal conditions. Also, the accuracy of the Image Method for adding artificial reverberation to anechoic speech recordings is established. A new algorithm, Pre-emphasised Maximum Likelihood Epoch Detection (PMLED), for Glottal Closure Instant detection is proposed. The algorithm is tested on natural speech and is shown to be both accurate and robust. Two techniques for giottai waveform estimation, Closed Phase Inverse Filtering (CPIF) and Iterative Adaptive Inverse Filtering (IAIF), are compared. In tandem with an LF model fitting procedure, both techniques display a high degree of accuracy However, IAIF is found to be slightly more robust. Based on these results, a Glottal Excited Linear Predictive (GELP) coding system for voiced speech is proposed and tested. Using a differential LF parameter quantisation scheme, the system achieves speech quality similar to that of U S Federal Standard 1016 CELP at a lower mean bit rate while incurring no extra delay.