Before it became the darling of circuit benders the world over, the Speak & Spell was a marvel of modern technology. Complete with a microprocessor and voice synthesizer, the Speak & Spell was able to speak a limited vocabulary that [Furrtek] thought should include words such as, “al qaeda”, “necrosis”, and “butt”. The Speak & Spell included an expansion port for cartridges containing a larger vocabulary, and with a huge amount of effort [Furrtek] created his own Speak & Spell carts that allow it to talk like a sailor.
The Speak & Spell ROMs were stored on a very strange memory chip; instead of a parallel or serial interface, the chip reads five nybbles at a time before returning the saved data. At first, [Furrtek] thought he could get an ATtiny microcontroller, but the way this memory chip is set up made it impossible to send and receive data even on a 400kHz I2C bus.
The project eventually found some decent hardware in the form of a CPLD-based cartridge that was more than fast enough to interface with the Speak & Spell. After that, it was only an issue of converting words into something the speech synth can understand with some old Windows 3.1 software and finally burning a ROM.
The end result is a Speak & Spell with a perverse vocabulary and is much, much more interesting than a circuit bent piece of hardware with a few wires crossed. Check out the video after the break.
[youtube=http://www.youtube.com/watch?v=JngcLfVQT5U&w=470]
Hello furrteck, you will find more modern tools for LPC and TMS5220 on this thread http://forum.system-cfg.com/viewtopic.php?f=25&t=3606
Excellent work researching and writing this all out! The LPC sounds could have used a little more love though, as they are currently unintelligible for the most part.
I agree, now the Speak & Spell can spell “Dog’s Bollocks” but can’t pronounce it. :)
I think it’s because words in the module are supposed to have two parts: the phonetic word and the actual word. The voice synth is based on phonemes, but the VFD shows the correctly spelled word. I recently picked up a S&S off eBay for $9 and am tearing it apart (has a bad transistor…someone plugged in a 9+ volt power source and zapped the power supply board – am in the process of fixing it, then on to building my own modules!)
Great hack – also a great reference for one of my own projects!
“CLPD”? *Really*, Brian?
Ridiculous pedantry? *Really*, Yarr?
no
Those TI speech chips were quite the marvel of their day. Unlike the voice addons for game consoles like Intellivision and Magnavox Odyssey^2/Phillips Videopac which used digitized recordings, TI’s chip could create genuine synthetic speech.
TI’s method is Linear Predictive Coding, which has a wide variety of parameters like pitch, attack, decay, sustain and more. A program written to allow the user access to all those could be used to create very human sounding speech, though it’d still sound like a low bitrate 8 bit recording.
Easier than that was to digitize a real person speaking then use a program (like Furrtek did) to analyze the recording and automatically create the LPC data to feed to the speech chip. That’s what TI did for games like Parsec and Moon Mine.
If you’ve played both Moon Mine and the DOS game Blake Stone: Aliens of Gold you might have noticed something very familiar in the DOS game. Dr. Pyrus Goldfire alternately says “You’ll never get me!” and “Ha ha ha ha!”, which sounds identical to the Zygonauts in Moon Mine. It would be quite interesting to extract the LPC codes from the TI game, run them through a Speech Synthesizer on a TI-99/4A to get a clean recording, and rip those sounds from Blake Stone – then do a side by side comparison.
I bet the DOS game’s authors recorded the Zygonaut’s voice from a TI computer and used the sounds in Blake Stone.
I kept a couple Speak ‘N Somethings in my parts box for years, thniking of using them to add speech to a project, but I finally decided I didn’t want to mess with the -19Volt chips.
thinking, NOT thniking, sorry! Where’s that edit button anyway?
Wait A 30 year old toy cannot be interfaced with an Arduino because the Arduino isn’t good enough?
[In best Speak’n’Spell voice] You are… correct!
Ardiuno is suck.
I still keep as a treasure my PCF8200 in the hope that some day will find some way to extract formants from digitized voice samples to make it speak. My SP0256-AL2 is the only one that talks to me.
I found a Speak-n-Spell at a garage sale that has its only fault a display showing words with broken segments. Are there parts available to replace the display (have to check ebay)? If not, I might try rebuilding mine with an Arduino and an LCD display from Maker Shed.
While I’ve never popped the hood on an S+S, dead segments on a VFD are sometimes due to bad solder joints caused by the pins getting too hot. A quick touch up with some good solder with plenty of flux will usually fix it.
Speak and Spell pic says “DILDO”
that’s what I noticed at first too, made me LOL pretty hard, haha. of all the words to use as a main pic, it had to be DILDO
Staying in the puerile spirit of this !! :)
Ya gotta wonder if Stephen Hawking
ever just pisses away a couple of hours
doing this with his speech device.
or imagine if the great minds that built
nuclear weapons had designed these instead
(in all the needed languages & dialects)
and We dropped them on each other,
what kind of world would we inhabit
as a result….
Remember at Showbiz Pizza back in the 80s where they had the Apple II with the educational software? Somebody left the machine with the type and speak program running. Soon it was saying every one of Carlin’s Dirty Seven. The next time I visited, they had locked out the Carlins.
Coudn’t understand any of the words it said to spell…
This hack is truly worth Hackaday. Great job!
Here is a link to an ebay sale (not mine) showing a word on a Speak-n-Spell that I’m not sure was on the actual one!
http://www.ebay.com/itm/VINTAGE-WORKING-SPEAK-AND-SPELL-TOY-BY-TEXAS-INSTRUMENTS-TI-SPEAK-SPELL-1978-/321034289101?pt=Educational_Toys_US&hash=item4abf227fcd
The only spoken word I could understand in that was necrosis, everything else was just unintelligible.
I’m sure that it must be possible to get the pronounciation better than that.
According to somebody who was trying to encode speech for a TMS5220 arcade game, TI had some pricey proprietary hardware ($25K) to do the work. While this isn’t the same as the TMS5100 in the S+S, it could provide some insight to the process. http://www.vectorlist.org/Vectorlist/2012/05/0062.html
IS THERE A WAY TO COVERT A SPEAK N SPELL FOR GHOST HUNTING
IS THERE A WAY TO CONVERT A BANANA FOR STOCK MARKET PREDICTIONS