A linear prediction vocoder for coding speech signals at 2.4 kilobits per second has been simulated on an HP-1000 computer system. The digitized speech signal is divided into frames of 20 ms, duration and for each frame, the voiced/unvoiced/silence decision, the pitch period if the frame is voiced and the gain of the autoregressive filter along with 10 reflection co-efficients are transmitted. Of the various analysis algorithms tested, the optimum performance is obtained using autocorrelation method for predictor co-efficient determination and inverse filtering plus Average Magnitude Difference Function algorithm for pitch extraction. Voiced/unvoiced/silence classification has been done independent of the pitch extraction algorithm to obtain a reliable classification. With these transmission parameters, the speech signal has been reconstructed using a digital speech production model.The performance of the vocoder has been evaluated using consonantal differentiation test for intelligibility measurement and...