SC-01A Speech Synthesizer

Last updated 12/04/07, dg

The Votrax SC-01A Speech Synthesizer is a phoneme synthesizer of the early 1980's.  It is capable of unlimited English speech using a stream of phoneme codes as input.  All transitions between phonemes are handled automatically.  (It thus stands in contrast to most other synthesizers of that time--see below.)  It was based on work by Richard Gagnon, and later was enhanced to become the SC-02, more widely known as the SSI-263.  This page has some resources that might be useful to the person studying or using the SC-01A.

(See the bottom of the page for a brief set of terms that might be useful.)

The speech it produces is of quite high quality for the time (and for a single-chip IC).  It rates reasonably well in intelligibility, but is not very natural sounding.  64 different phonemes are available, and 4 levels of intonation.

The SC-01A (nor SC-01) is not available on a regular basis from any known suppliers.  The chip was last made in the late 1980's, and SSI told us that due to the types of processes used in this analog chip, it was unlikely to be made again--too complicated to transfer the technology.  (However, we never had really good communication with SSI, so take this with a grain of salt.)

(2/24/04) Update!  A posting to rec.games.vectrex claims that a stock of SC-01's can be found at http://www.best-electronics-ca.com/custom-i.htm.  They do show a listing for "CO61720        (1400 / 1450 Speech SC01)        $30.00".  Their main business is to supply Atari parts.  According to someone who called, they have maybe 20 left.

Brief comments on the SC-02 / SSI-263 / "Artic 263"

The SSI-263 (or whatever you prefer calling it) is quite different from the SC-01; it has many more control registers, a different package and pinout, and so on.  As far as I know, it has the same analog formant synthesis core as the SC-01, and is also capable of intelligible (but not necessarily natural sounding) speech.

(July 2000)  In July 2000 I heard of two companies that claimed to still have access to stocks of the 263--very interesting.  The first is Artic Technologies ((248) 588-7370, Troy Michigan, http://www.artictech.com/).  Their website implies the chip is still in production, but a sales person indicated that the last buy on the chip was a couple of years ago, around 1998.  (This conflicts with information I had back in the early 1990's, but as I said, information did not flow very well back then.)  SSI was not the manufacturer at that point (Semtech might have been at one point, and it may have passed to yet another house by the end).  Artic has a stockpile of "Artic 263" chips for their own products, and is willing to sell chips.  A ball park on the pricing is around $100 qty 1, $90 qty 10+, and they can quote on larger quantities.  Interesting note--Artic founders were involved in the development of the Votrax chips (http://www.artictech.com/history.htm).  (As of mid-2002, Artic is still quoting about $100/chip.)  The second company I heard about through a personal e-mail.  This person seemed to indicate their company had a steady supply of 263's, but this person wasn''t able to discuss any details, and in any case it didn't sound like any chips were available to sell to anyone else.  Quite possible that their company was also running on a stockpile from a last buy, but who knows.

(10/20/05) Another SSI-263 Source! Henry Courbis of GSE-Reactive tells me that he has some SSI-263's for sale, $50 each. http://www.gse-reactive.com They also have various Apple IIe and IIgs parts (including a clone of the Mockingboard).

(4/11/06)  See below for SSI-263A datasheet and user guide.

The bottom line?  Single-chip analog formant synthesis is likely down for the count.  There are some recent offerings, though, that could fill the niche of single-chip phoneme synthesizer.  First is the SpeakJet, a pre-programmed PIC chip that takes serial data in (phonemes, though probably not the same ones as the SC-01 or 263) and outputs PWM speech.  Second is the Winbond text-to-speech chip.  For details on both, see below under Modern Alternatives.

Data Sheets

I scanned in a copy of the SC-01 data sheet.  It is in PDF format, but in fact it is entirely bit-mapped graphics (i.e., it has not been converted to actual text).  The resolution is 300 DPI.  It is readable for the most part, though some small heavy fonts are hard to read.  By the way, after running into a dead-end trying to get permission from Votrax to post this data sheet (they sent me to Artic Technologies, also in the Detroit area, who never got back to me), I noticed the explicit permission granted on the first page to reproduce this data sheet.  So feel free to redistribute as you like.

Update--12/07:  Bob Grieb was kind enough to furnish an excellent scanned copy of the datasheet.  It is scanned at 600 dpi, I think, and the original is loads better than anything else I've seen.  I've kept the old copies here as well.

Raymond Weisling kindly forwarded a scan of a better copy of an SC-01 datasheet, along with SSI-263A data sheet and user guide.  (Will post pending permission from T.I.)  This is saved in JPEG format, and I've also packaged it as a PDF, but again the PDF is just a wrapper around graphic images.  I've also included links to the jpg files themselves in zip files if you'd prefer to have them that way.  (By the way, check out Raymond's Four Letter Word and other Nixie-tube creations at:  http://www.zetalink.biz/produ.html)

Files:
Recommended:  SC-01 datasheet, 9 pages, good original, PDF:  sc01.pdf (588kB)
SC-01 datasheet, 9 pages, good original, 150 dpi/JPEG.  PDF:  sc01jpeg.pdf (829kB)  JPEG/ZIP:  sc01jpeg.zip (805kB)
SC-01 datasheet, 10 pages (last is application circuits), poor original, 300 dpi/TIFF. PDF:  Poor10PageSc01.pdf (627kB)

If you have an SC-01A datasheet, or related material that could be posted, please let me know.

Phonetic Speech Dictionary for the SC-01 Speech Synthesizer:  Dave of www.riana.com has posted a PDF of the little dictionary showing SC-01 phoneme sequences for some common words:  http://www.riana.com/electronics/sc01/index.html

SC-01 vs. SC-01A

What is the difference?  Klatt's review paper below states there is a quality difference between the two.  Can anyone go into more detail on the differences?  (See below for my email if you have any information.)

What happened to SSI?

I'm a little unsure, but it sounds like Texas Instruments acquired a lot of SSI's assets in 1996, according to this press release:  http://www.ti.com/corp/docs/press/company/1998/98005.shtml
So some SSI elements became part of TI's Storage Products Group (http://www.ti.com/sc/docs/products/storage/index.htm).
However, I was in touch (April 2006) with that TI group, and they did not have any information on the 263 (or other chips).  They thought that Teridian Semiconductor Corporation (http://www.tsc.tdk.com/) might have retained the speech products.  However, my email to them remains unanswered.  Anyone have additional information?

Resources

For the theory of operation, see the following patents:
Voice Synthesizer, Richard Gagnon, 3,908,085, 9/23/75
Speech Synthesizer Responsive to a Digital Command Input, Richard Gagnon, 3,836,717, 9/17/74
Voice Synthesizer, Mark Dorais (assigned Federal Screw Works (=Votrax)), 4,128,737,12/5/78
Voice Syntehsizer, Carl Ostrowski (assigned Federal Screw Works), 4,130,730, 12/19/78
(I recommend Google's patent interface, but the USPTO now has access to patents online.   www.uspto.gov)

See also Gagnon, R. T. (1978).  "Votrax Real Time Hardware for Phoneme Synthesis of Speech," Proc. Int. Conf. Acoust. Speech Signal Process. ICASSP-78, 175-178.  (Many thanks to Eric Smith for finding this citation!)

From "Talking Terminals," David M. Stoffel, Byte, September, 1982:  "The Votrax VSA and VSB synthesizers seem quite similar with respect to their phoneme production, but the FSST-3, which uses the VSA, definitely sounds inferior; whether this is an artifact of the VSA synthesizer or poor audio amplification, I don't know. You may wonder why none of these products uses the new Votrax SC-1A (sic) integrated circuit, which is less expensive. The single quantity cost of the VSB is about $800, while the SC-01A is $70. But there are two major reasons why the SC-01A is not used. The speech-rate and pitch controls are both dependent on the same clock signal or timing circuit, affecting the ease with which intelligible speech may be produced. Also some people are concerned about the acceptability of the SC-1A's (sic) sound quality. Only scientific performance measures can determine which Votrax synthesizer is ultimately more intelligible. (For a description of an application using the Votrax SC-01A speech-synthesizer chip see Steve Ciarcia's article on page 64 in this issue.)"  See http://www.lindenreport.com/stoffel/talk.html

For the best overview of speech synthesis up to the late 80's, see
Klatt, Dennis, "Review of Text-to-Speech Conversion for English," Journal of the Acoustical Society of America, 82:3, September 1987, p 737-793.  This is an excellent overview, and includes (pg 756) a brief description of the SC-01 that starts as follows::
"Apparently oblivious to all of the prior research detailed earlier, a man experimenting in his basement workshop, Richard Gagnon, designed a synthesis-by-rule program that eventually resulted in the Votrax SC-01 chip. ... It is a remarkable device for the price."
No mention of Mozer (of National Semiconductor's speech synthesizer, if I'm remembering correctly), but this is the best article to start with if you want an introduction to the synthesis of speech.
Important note:  Klatt's paper is now online!  See this and other papers at http://www.mindspring.com/~ssshp/ssshp_cd/ss_home.htm

Prochnow, Dave, Chip Talk: Projects in Speech Synthesis, Tab Books, Blue Ridge Summit, PA: 1987.  ISBN is 0-8306-1912-7 (hard cover) and 0-8306-2812-6 (paperback).  Hardcover is Tab # 2812.
A hobbyist-oriented book on speech synthesizers circa 1987, including

as well as stand-alone systems.  Schematics, pin-outs, construction hints.  Not much detail on theory.  All but certainly out of print, but that's what inter-library loans are for!

For a deep account of one text-to-speech system (the basis for some of the best speech synthesizers until perhaps recently), see the book From Text to Speech: The MITalk System, by Allen, Hunnicutt, and Klatt, Cambridge University Press, 1987.  (The pseudo-code in the back is not, however, without a number of errors and omissions.)  The parameter-to-speech part of MITalk is detailed for the most part in Klatt's "Software for a Cascade/Parallel Formant Synthesizer," J. Acoust. Soc. Am., 67:3, March 1980, pg 971-995.

See http://www.mindspring.com/~ssshp/ssshp_cd/ss_home.htm, the Smithsonian Speech Synthesis History Project, which includes audio of a variety of synthesizers, the Klatt paper, and personal recollections.  See especially the chronology of Votrax's speech products.

Obviously, newsgroups such as comp.dsp, comp.speech.*, comp.arch.embedded, and comp.robotics.*--and their associated FAQ's--are valuable resources.

See also http://www.robotprojects.com/voice/voice.htm, by Scott Savage, which has some interesting links on speech synthesizers.

(7/2/02)  Tom McClintock notes the following:  "One item of interest regarding the SC-01a. The 'PinMAME' developers have incorporated SC-01a emulation into their pinball simulations. The source code includes digital representations of all the phonemes. Pretty cool stuff, but complete and accurate emulation is not quite there. Check out the source: http://pinmame.retrogames.com/release/pinmame_112_1_src.zip"

Bob Paddock wrote up a nice list of links at http://www.chipcenter.com/circuitcellar/june00/c0600rp42.htm
 

Applications

Ciarcia, Steve, "Build a Low-Cost Speech Synthesizer Interface," Byte, June 1981, p 46.
Ciarcia, Steve, "Build an Unlimited-Vocabulary Speech Synthesizer," Byte, September 1981, p 38.
Ciarcia, Steve, "Build the Microvox Text-to-Speech Synthesizer, Part 1," Byte, September 1982, p 64.
Ciarcia, Steve, "Build the Microvox Text-to-Speech Synthesizer, Part 2," Byte, October 1982, p 40.
(See http://members.tripod.com/werdav/t2smicrv.html for the article above.)
Ciarcia, Steve, "Talk to Me:  Add a Voice to Your Computer for $35," Byte, June 1978, p 35.
Ciarcia, Steve, "Build a Third-Generation Phonetic Speech Synthesizer," Byte, March, 1984, p 28.  (SSI-263)
Note that some of these articles are collected in "Ciarcia's Circuit Cellar" volumes I, II, and III.  Volume I covers 9/77-11/78, II covers 12/78-6/80, and III covers 7/80-12/81.
Technical details of Gottlieb pinball machines, some of which use an SC-01A chip:  http://www.ionpool.net/arcade/gottlieb/technical/sound_boards.html
The  Type N Talk manual:  http://members.tripod.com/werdav/txtospm1.html
Intex Talker (uses SC-01A):  http://web.inter.nl.net/hcc/davies/txtospin.html
 

Where it was used

One approach to finding SC-01A's is to buy used devices that have this chip in them (e.g., eBay).  The following include an SC-01A: (what others am I missing?)

Brief comments on other single-chip synthesizers of the past:

The late 70's to mid 80's saw a number of speech synthesis chips developed.  Below are brief comments to place some other chips of that era in relation to the SC-01.
Texas Instruments TMS5100, TMS5220, etc.  TI developed a series of synthesizers using LPC (linear predictive coding).  LPC is a compact method of encoding speech, and so typically systems using these chips have a limited, pre-specified vocabulary, though in a few instances software and lots of data was used to create unlimited speech systems (e.g., Street Electronics).  (LPC is still used as the core of most digital speech compression algorithms, including for digital cell phones.)
General Instruments also produced some LPC-based synthesis chips.  (And one chip that supported LPC analysis for speech recognition--the SP1000?  Ciarcia had an article on the chip.)
General Instruments SP0256-AL2.  According to Sclater (Neil Sclater, Introduction to Electronic Speech Synthesis, 1983, Howard W. Sams), a digital formant synthesizer using a serial data stream from special ROMs.  Can be paired with the CTS256A-AL2, which is a hard-coded PIC7041 microcontroller (this, according to Prochnow) that has a built-in text-to-speech algorithm and ROM with allophones.  (Is there any transitioning of parameters by the built-in program?  Or are allophones just concatenated, which would drastically reduce the quality of the speech?  I don't know.)  Update!  See http://www.primenet.com/~im14u2c/intv/tech/ivoice.html for a very detailed description of the SP0256 along with C source to simulate it, by Joe Zbiciak.  The SP-0256 is in fact a 12-pole LPC synthesizer and not a formant synthesizer.  It has a simple microsequencer that can execute a handful of instructions.  (Thanks to Eric Smith for the link!)
Philips PCF8200.  A digital formant synthesizer, it requires a constant flow of parameters to synthesize speech.  Typically these parameters were derived from actual speech, but in theory, you could create these parameters using software (such as Klatt's algorithms) to provide unlimited speech.  (Formant parameters are "easily"--and directly--synthesizeable from abstract rules; LPC parameters are not easily directly synthesizeable from sets of rules.)  Essentially, a digital version of the formant filters in an SC-01, but without the transitioning logic found in the SC-01 (such transitions in the SC-01 were generated using analog circuitry).
National Semiconductor DigiTalker (MM54104).  A direct waveform encoding/decoding chip set.  Uses ROMs with a limited number of words.
 

Modern alternatives

1.  Voice recording chips

Currently available is a line of voice recording chips from ISD (now Winbond).  These are even available at Radio Shack and Digikey.  However, they are waveform recording devices, so not capable of unlimited speech.

2.  Single-chip phoneme synthesizer

(2/22/04)  A new pre-programmed PIC that does single chip speech synthesis and sound effects--the SpeakJet.  Apparently released within the past two weeks, it accepts serial date in (phonemes) and output a PWM signal that with minimal (2-pole) filtering can be fed to an amplifier and then speaker.  Internal oscillator.  Seems to run about $25.  Developed apparently by Magnevation (www.magnevation.com) and Scott Savage (oopic.com) over the past 5 years.  I have only heard a few demos.  Widely available through robotic supply sources.  The interface and command set look very well thought out.  This might turn out to be a very nice chip for applications that would normally want to use the SC-01A.  According to an email I got from Scott, the SpeakJet does do transitioning between phonemes.  If anyone has additional details how this works, what PIC it is (someone guessed an PIC18F1320), etc., let me know.  (This conjecture makes sense--the 18F1320 has an 8x8 mulitplier, 8k bytes of program space, PWM, and runs about 10 MIPs.  This is more than enough to do a stripped-down digital formant synthesizer.  A full bore, unoptimized KLATTalk-ish formant synthesizer core will run on a 10 MIPs 16-bit wide chip with MAC.)

Update (12/4/07):  Robert Doerr (http://www.robotworkshop.com/) just wrote an article in the December issue of Servo Magazine (http://www.servomagazine.com/) about using a small microcontroller to translate SC-01 phonemes into SpeakJet allophones, plus handle the interface signals so you can plug the circuit into a regular SC-01 22-pin socket.  If you need to replace an SC-01, but keep the rest of your circuitry intact (e.g., Hero robot), this could be an interesting solution.

Also, Chip Gracey, Parallax founder and the designer of the Propeller chip, has apparently been working on speech synthesis that would run on the Propeller.  (See Make magazine volume 10:  http://www.make-digital.com/make/vol10/?pg=78&search=parallax+propeller+speech&u1=texterity&cookies=1.)  Anyone with additional information?  If it ran on just one of the eight 32-bit processors (which should be quite realistic), this would be interesting for new embedded applications.

3.  Single-chip text-to-speech

(7/2/02)  Robert Doerr points out a newish chip from Winbond, the WTS701, which includes text-to-speech algorithms.  http://www.winbond.com/E-WINBONDHTM/partner/b_2_a_5.htm. 

(5/14/03) Tom Arnold points out that the datasheet is finally available for the WTS701, along with a live demo (you type in text, get back audio output)  From the description, it sounds like it stores speech (using the ISD technology) on chip, concatenating to form the output.  (See their FAQ on the page above.)  Surface mount package.  SPI interface.  (See sources below for a module.)

4.  Do it yourself DSP

You could also port open-source speech synthesizers to a microcontroller/DSP platform.  Not trivial--you'll need on the order of 5-10 MIPS of 16-bit wide processing (with MAC), and the digital-to-analog output along with at least 32-64 kB of program space plus some RAM.  Probably not very economical given the options above, but very educational if you are into that sort of thing.

As for speech synthesis by concatenation ("why not just record all 64 sounds from the SC-01 and string them together as you want?"), see the comment at the bottom of this page.

Sources for various chips

Please note that I do not have any SC-01A's or 263's for sale.

(2/24/04) Update!  A posting to rec.games.vectrex claims that a stock of SC-01's can be found at http://www.best-electronics-ca.com/custom-i.htm.  They do show a listing for "CO61720        (1400 / 1450 Speech SC01)        $30.00".  They supply Atari parts.  Apparently about 20 left.

(10/20/05) See the comments under SSI-263 at the top for a new source for the SSI-263.

(2/22/04)  A source for some speech chips (not including the SC-01A):  http://www.speechchips.com/shop/  The sell the SpeakJet, SP0256-AL2, CTS256-AL2, and even the SP1000!

(2/22/04)  There's a UK company selling a module with the Winbond WTS701 (at least that's what it looks like):  http://www.totalrobots.com/access_files/speech.htm  40 pounds UK (around $75).

A few useful terms:

Formant:  A peak in the spectrum of speech corresponding to a resonance in the vocal tract.  A formant synthesizer uses bandpass filters, typically, to create these resonances.  Depending on the type of speech to be synthesized (e.g., female, male), four or usually five or more formants are necessary for reasonable speech.

Phoneme:  An abstract sound unit, for example a sound like "eh".  The problem is that the actual sound associated with a phoneme depends on its context--the "eh" sound is influenced by what sounds preceed and follow it, for example.  The SSI-263 has some 60 phonemes it uses to synthesize English speech.

Allophone:  The actual realization of a phoneme (i.e., in a particular context).  More than one allophone can be associated with a given phoneme.
 

One more comment...

One idea that seems to come up when discussing the SC-01A (specifically, the lack of availability), is that of creating a software emulation by recording the 60+ phonemes and just concatenating them.  It turns out that much of the intelligibility of speech is wrapped up in the transitions between the ideal phoneme sounds ("targets"), so the concatenated speech will sound no where near as good as the SC-01, which has internal circuitry to generate transitions.  In fact, there's a speech synthesis method based on just concatenating the transitions between phonemes, called diphone synthesis.

Feel free to e-mail me if you have any interesting resources or note errors on this page:  dgrover at redcedar.com