A programmable digital signal processor (PDSP) intended primarily for speech synthesis applications has been developed using custom NMOS LSI circuitry. Key features include high speech quality, great flexibility, simple interfacing to standard microcomputers, low cost, low power, small size, and the stability of a digital processor. Because of extensive programmability, substantial computational power, and a selection of hardware configurations, the PDSP components can implement a wide range of vocal tract model structures at costs roughly proportional to performance. For example, the user can create a cascade formant synthesizer, a parallel formant synthesizer, a hybrid cascade/parallel formant synthesizer, or a lattice filter structure. Typical applications encompass fixed‐vocabulary speech output devices, systems for translating unrestricted text into speech, and low‐bitrate/secure voice communications. The custom devices are currently being designed into a high‐performance speech output reading system for the blind.