For many people, their first glimpse of the power of technology wasn’t a laptop or a cell phone. It was a bright orange device that spoke in an odd, rattling voice: The Texas Instruments Speak & Spell.
First announced at the CES show in 1978, the Speak & Spell was an educational toy that offered a number of word games, such as a version of hangman, guess the word and a spelling test. It included a keyboard, one line display (a vacuum fluorescent display) and a speaker tied into a speech synthesizer.
This was revolutionary because, rather than playback a pre-recorded voice, it could synthesize speech. Earlier toys like the Chatty Cathy doll of the 1960s had played back pre-recorded voices stored on plastic discs, but the Speak & Spell used a revolutionary voice synthesis system that electronically simulated the human voice. This system was designed by engineers at Texas Instruments as an offshoot of their pioneering research into speech synthesis and, according to a Texas Instruments press release, the engineers (including lead engineer Paul Breedlove), brainstormed ideas on how to use this new technology and came up with the $50 toy.
“The challenge was that it had to be solid-state (no pull strings!), cheap (meaning it used a low-cost semiconductor technology), and the speech had to be good enough so that the user could understand the word out of context ” said Richard Wiggins, one of the designers in a 30th anniversary interview with Vintage Computing & Gaming.
The Speak & Spell marks the first commercial use of the new concept of Digital Signal Processing (DSP), where a silicon chip digitally processes a signal to shape and create the voice, and was, according to the company, the first time that “the human vocal tract has been electronically duplicated on a single chip of silicon”. The Speak & Spell was built around a TMC0280/TMS5100 speech synthesis chip and a TMS6100 memory chip that could hold 128kbits of data.
The memory chip held the details of the components of words (called phonemes), and a series of rules on how to use this data. The DSP speech synthesis chip used this data to create a voice from the text. This large amount of data could be stored in the small memory of the device using a method called Linear Predictive Coding. “We generally stored strings of data that corresponded to a particular word” recalled Wiggins. “Patterns for new words were generated back in the laboratory — sometimes by experimenting with data, but certainly the best results were obtained by large computer programs that analyzed real speech data and determined data patterns and strings of data that, when applied to the synthesizer, resulted in words that could be easily understood and sounded very much like the original speech.”
Together, this compression technique and the DSP could create a synthesized voice that could synthesize a spoken word rather than just playing back a recording. That might sound like a trivial difference, but it paved the way for every device since that has been able to speak, from Siri to the high-end artificial voice software behind some Japanese pop stars.
Of course, the Speak & Spell had limitations, including a small 200-word vocabulary and a rather odd American accent. Both of these could be overcome (to some degree) by the addition of ROM packs, which expanded the vocabulary and added different languages. Versions for the UK, Germany and France were also made, complete with the appropriate language components.
The chip inside the Speak & Spell was also used in other devices: the speech synthesis module for the BBC Micro was built around the same TI chip. It was also used in a number of arcade games and Texas Instruments own TI 99/4A home computer.
Several iterations of the device followed in subsequent years, with the Speak & Spell Compact in 1982 and the Super Speak & Spell in 1992, which included an LCD screen. This was the final version, though; the increasing power of personal computers meant that speech synthesis tasks could be handled by software, which was a much more flexible (if not always comprehensible) approach.
The Speak & Spell Today
The Speak & Spell lives on, though, finding a home in a number of surprising locations.
The simple design of the Speak and Spell has made it popular with circuit benders, who take technology and hack it to do strange and unusual things. They have a particular fondness for this gadget because the circuit board has lots of points to tap into it, making altering and replacing the signals flowing around it much easier. These musical hackers have done a lot of interesting things with the Speak & Spell, including adding switches that control the pitch, speed and distortion of the voice, and modifying it to receive MIDI signals, so it can be triggered by the universal language of the musical world.
More conventional musicians have also found a friend in the Speak & Tell, with the unmistakable wobbly voice and bob-bip-da-beep start up sound appearing in songs by Kraftwerk, Coldplay and many others. Depeche Mode even named an album after the device. The Speak & Spell also had an important role in the film E.T., where one formed part of the device the alien hero used to phone home.
This longevity is why the Speak & Spell is a gadget we miss. Although it was a cheap kids toy, it became much more than that. It became a pioneering piece of technology that still retains a place in many peoples hearts, and was the gateway drug for a generation of hackers, tinkerers and curious people who went on to build the next generation of technology. Long after Cabbage Patch Kids were consigned to the great toybox in the sky, the Speak & Spell was still finding new ways to inspire people and showing up in unexpected place.
Are you nostalgic for the Speak & Spell? You can try it out thanks to the emulator created by Connecticut science teacher Kevin St. Onge here, or they show up on eBay quite often. For those looking to create their own Speak & Spell inspired projects, Onge also offers a download of the audio files from the device. For more information on the various models of Speak & Spell and the internals of the device, DataMath is the best place to start.