An automatic public address system using a PARCOR speech synthesizer LSI has been developed. A µCPU controls the synthesizer LSI, and compiles up to 255 sentences from maximum 255 words. For the excitation source, a one-pitch residual waveform is used to improve the PARCOR-synthesized speech sound quality good for the public address system. Adopting the differential coding of the K parameters, in the case that the frame period is 10 ms and the spectral distortion per parameter is 0.5 dB, spectral information has been reduced 26.9% as compared with that of the conventional PCM coding of the K parameters. Furthermore, varing the frame data length, the total data have been reduced 30 to 35% as compared with the fixed frame data length. By means of the spectral distortion measure, spectral information has been compressed in this variable frame rate system. Consequently, is has been known that the speech sound quality is improved by the parameter smoothing effect. An average bit rate is 4.2 kbits/s (in case of 10 ms frame period) or 7.2 kbits/s (5 ms frame period). Sound articulation score is 94.4% (10 ms frame period). Finally, a general method to derive the excitation waveform from the mean power spectrum of original residual signal is proposed, and it is shown that the use of novel excitation has the effect of synthesizing a similar sound to the original one.