Speak & Spell

A spelling toy that had to talk

The Speak & Spell began as a difficult proposal inside Texas Instruments. Speech synthesis in the mid 1970s belonged to laboratories, and the idea was to shrink it into a red plastic device sold to children. Paul Breedlove led the small team that did it. Development started in 1976 with an initial budget of $25,000, as an outgrowth of TI's speech research.

The IEEE record fills in the dates. In December 1976 Richard Wiggins demonstrated the concept to Breedlove, Larry Brantingham and Gene Frantz in TI's Dallas research laboratory, and the group had a working product by April 1978. The toy was introduced at the summer Consumer Electronics Show that year, advertised for children aged seven and up, with a list of more than 200 commonly misspelled words.

The shipping version, the Basic Unit, needed no cartridge: it carried Mystery Word, an electronic hangman, plus Secret Code and Letter, and showed eight characters at a time on a vacuum fluorescent display. That limit is why the machine speaks in fragments rather than sentences.

One chip doing the work of a rack

The component that made the toy possible was the TMC0280, later sold as the TMS5100. It generated speech using tenth-order linear predictive coding, and IEEE describes it as the industry's first digital signal processing integrated circuit for speech generation, the starting point of the consumer DSP market. Wikipedia goes further and calls it the first single-chip voice synthesizer.

Linear predictive coding does not store a recording. It stores a set of coefficients that model the vocal tract, and the chip rebuilds the sound from them. The saving is enormous: the Speak & Spell kept its words at a data rate of roughly 1,000 bits per second, far below what sampled audio of the same length would need. The phoneme data sat in a pair of 128 Kbit metal gate PMOS ROMs, at the time the largest capacity ROM in use. Rebuilt from a handful of parameters, the voice loses the breath of a recording and gains a buzz, and that buzz became the product's signature.

Recording a voice, then taking it apart

The words started as a human performance. A professional speaker read the list, the utterances were captured and processed into parameters, and the American voice came from Mitch Carr, a radio anchor at KRLD in Dallas. Localisation moved the work to Europe: by 1982, with British, French, Italian and German versions in preparation, voices were recorded at a TI facility near Nice in France and the full bit rate recordings were sent to Dallas. Weeks later the reduced data came back and needed heavy hand editing. The rate had been cut so aggressively that some words were unintelligible and took days to repair, and others were scrapped.

Cartridges in the battery bay

The Speak & Spell was also a handheld computer with a library slot. Expansion modules plugged into a compartment beside the batteries, and a key on the keyboard selected them. Vowel Power arrived in 1978, with Super Stumpers, Basic Builders and Mighty Verbs following in 1979. Because the modules were plain ROM libraries with no regional lockout, the 1980 redesign, which replaced the raised buttons with a membrane keyboard, still ran the original library. Wikipedia counts the toy among the earliest handheld devices with a visual display to take interchangeable cartridges.

TI then built a family on the same shell. The Speak & Read came in 1980 for ages four to eight, and the Speak & Math the same year for ages six to twelve. The line was sold in the United States, Britain and Japan and localised into at least nine countries and seven language variations. The German version was the Buddy, the French one La Dictée Magique.

What it left behind

Two things outlived the toy. The first was the silicon. A variant of the speech chip appeared in Chrysler cars during the 1980s as the Electronic Voice Alert, and TI adapted the technology into a speech accessory for its TI-99/4A computer. IEEE dedicated a milestone plaque in Dallas on 15 October 2009, calling the toy the start of a digital signal processing industry it then valued at more than $20 billion.

The second was cultural. A 1978 model appears in E.T. the Extra-Terrestrial in 1982, wired into the alien's improvised communicator, and TI shipped an E.T. Fantasy Module the same year. Kraftwerk sampled the opening sequence; musicians opened the cases and cross-wired the keyboard matrix, which became part of circuit bending. Units sit in the Computer History Museum and the Smithsonian.

Production ran from 1978 to 1992. The Super Speak & Spell of 1989 replaced the vacuum fluorescent display with a 16-character LCD and a QWERTY keyboard, and the last version, for Spain, was a renamed El Loro Parlanchín. In 2019 Basic Fun revived the second-generation design with a newly recorded voice processed to sound synthesized, a segmented LCD, and no cartridge slot. The machine no longer generated speech. It imitated a machine that had.

What the record leaves open

The engineering is well documented; the commercial record is not. No reliable sales total turned up in the sources read here, and the launch price could not be confirmed, so neither is given. What the record does support is a $25,000 research allowance, four named engineers and a chip that outgrew the toy it was built for.