ESS Technology - Professor Mozer's Patented Technology

Professor Mozer's Patented Technology

Professor Mozer first became interested in speech technology when a blind student in his class in 1970 asked whether he could help design a talking calculator. Mozer spent 5 years working on it, and his speech technology first appeared in the Telesensory Systems "Speech+" talking calculator, in a chip called the "CRC Chip", more commonly known as s14001a, the first self-contained speech synthesizer chip. This chip was also used in a few arcade games, notably Atari's Wolf Pack, and Stern Electronics' Berzerk and Frenzy, and in several of Stern's pinball machines.

After a three-year exclusive deal with Telesensory Systems from 1975 to 1978, Forrest Mozer sold a 3 year license to National Semiconductor, and they created another chip using Mozer synthesis, the MM54101 "Digitalker". At first, even then, all words were encoded by hand by Mozer in his basement, but in the third or fourth year of the license, National came up with a software encoder for it. After the exclusive license expired (National seemed to have a "non-exclusive" license for a year or so), Mozer licensed the technology to ESS. After Mozer's son Todd split off and created Sensory Circuits Inc., the technology was licensed there.

According to the Sensory Inc. history pages and old datasheets, they offered three types of compression:

• MX (this compression is nearly identical to that used on the Digitalker, with some minor coding changes and possibly some RLE. It's apparently used on some alarm systems and on the Vtech talking baseball/football cards)
• CX
• SX

and a few other PCM/LPC based systems.

Later, Sensory bought up the remains of many old speech synthesis manufacturers from the 1980s, such as Texas Instruments' speech division.

Professor Mozer's technique not only produced very realistic sounding speech, it also required very little on-chip (later, in software) RAM, a sparse and expensive commodity at that time. The advanced compression algorhithm (patented, an early form of psychoacoustic compression using similar spectra of ADPCM-encoded waves) reduced the memory footprint of speech about a hundredfold, so one second of speech would require 90 to 625 bytes. With ESS-speech, samples that would normally require almost all of the 64 kilobyte memory of the Commodore 64 (if encoded in PCM) were so small, that the entire game fit into the RAM along with speech, without requiring additional loads from disk.