One of my hobbies is collecting, restoring, and using speech synthesizers from the 1980s. In this page I’m going to catalog several of the different speech synthesizers that I have used and summarize them in one page. I’ve accumulated enough of these things that it’s hard to keep track of them all!
Phonetic Speech Synthesizers
Phonetic speech synthesizers break speech down into a series of phonemes, which are the basic building blocks of speech. For example, my name, “Scott” can be broken down into the Votrax SC-02 phones S/K/A/T, with the “A” in this example being the “ah” sound. There’s not always a 1:1 pairing between English language letters and phonemes. For example there’s often lots of different A sounds: A, AI, AE, AE1, AH, AH1, AW. Phonetic speech synthesizers have infinite vocabulary because you assemble these basic building blocks to form many different words.
The SP0256A-AL2 was a popular chip because you could walk down and buy it right there at your local Radio Shack store complete with data sheet. The IC supports 64 phonemes, but has no support for inflection, so you tend to get very monotone output from it. In my opinion it sounds alright but not great. The SP0256A-AL2 does have a companion IC, the CTS256A-AL2 that offers a full text-to-speech capability (though it does require several additional ICs to do so).
- CTS256A-AL2 Text-to-Speech Board.
- ISA (IBM PC) Speech Synthesizer Board using SP0256A-AL2.
- Speech Synthesizer for the Heathkit H8.
- Z8000 Speech Synthesizer.
- Speech and Sound Boards for the Epson QX-10 Computer.
- Speech Synthesizer for the RC2014.
- Speech Synthesizer for the Olivetti M20.
- Jupiter Ace Speech Synthesizer.
- Color Computer Speech and Sound Pack
- Beware of Counterfeit SP0256A-AL2 chips.
As you can see above, pretty much every time I get a new retro computer, I end up (re)implementing an SP0256A-AL2 speech synthesizer for it. The ICs are still readily available, and the circuit is pretty simple.
Votrax SC-01 / SC-01a
The SC-01 sounds better than the SP0256A-AL2, in my opinion. Part of this may be that in addition to having 64 phonemes, it also has four levels of inflection, which allow you to add additional nuance to the generated speech. Interfacing the chip is complicated in that the two inflection bits are interfaced separately from the 6 data bits for the phonemes.
- Colorware RealTalker for the TRS-80 Color Computer
- Heathkit Hero-1 Robot
- Hearhkit Hero Jr Robot. My modern rebuild of the Jr (video).
Votrax SC-02 / SSI-263 / Artic 263 / Aicom AI901
These all appear to be the same IC, but with different branding. The SC-02 adds several features above and beyond the SC-01, including control of Articulation, Inflection/Pitch, Filter Frequency, Rate, and Amplitude. It supports 64 phonemes. It’s interfaced like a traditional 8-bit peripheral with chip select lines, register select / address lines, and an 8-bit data bus.
Sampled Word Speech Synthesizers
These speech synthesizers typically sample whole words, compress them with algorithms such as LPC, and store the compressed sampled speech in a word-by-word basis. They can often yield higher quality speech than phonetic speech synthesizers, but at the cost of limited vocabulary. You can only have as many words as you took the time to sample, and as you have enough ROM or other storage to store them.
National Semiconductor Digitalker
The Digitalker was designed by computer speech pioneer Forrest Mozer. Back in the 80s you could find the Digitalker chip set at such places as Jameco or Digikey. Speech was typically stored in ROMs. You give an address to the Digitalker chip, and it in turn retrieves the right word from the ROM and plays it. The vocabulary is limited to what you can fit in ROM, though typically you could add as many ROMs as you like using bank switching techniques. Digitalker audio really sounds pretty good. The software for creating new Digitalker ROMs has been located, but the software and tools for compressing the speech has not. Thus if you have an existing vocabulary file, you can use the software to select words from it to generate ROM images, but there’s no software for recording your voice to create that initial vocabulary file.
Texas Instruments TMS5220
The TMS5220 is an LPC (linear predictive coding) speech synthesizer. The TMS5220 accepts 8-bit data through a typical data bus interface and requires an external processor to continuously stream this data to it to generate speech. In some cases, this processor could be the computer’s main processor, or a speech synthesizer implementation might make use of a dedicated CPU, for example a Z8, to serve as a dedicated speech controller.
- HP27201A Speech Output Module.
- TI 99/4A Speech Synthesizer