Voice Synthesis Chip SP0256-AL2 SOURCE

erco

I can't wait any longer for a Prop-based voice synthesizer, so I'm taking a giant leap backwards to 1998. See Jon Williams article at http://www.parallax.com/dl/docs/cols/nv/vol1/col/nv40.pdf . Other than locating that obsolete chip, he makes it look pretty dang easy, even with a BS1.

The chips appear occasionally on Ebay for high prices (there are currently some SP0250 chips for $30 that are much harder to use). I tracked down a UK source for the SP0256-AL2 chip at http://sciencestore.co.uk/acatalog/Electronics.html (bottom page) for GBP 9.50, or about $14.63 by today's rates. THAT is quite a bargain, my friends. Shipping was high, so I ordered ten to subsidize. That's four more than I need, so I'll part with four of them. First shot goes to my Parallax forum pals, if no one want them, I'll list them on Ebay and probably won't have any trouble selling them.

PM me if interested.

blittled

You wouldn't happen to have the crystals that go with these or know of a source? I bought the SPO256 back in the days when RS sold it but being a teenager scraping lawn mower money together I didn't buy the crystal. I keep kicking myself for not getting it. I have used a color burst crystal and that works well but I'm a purist and like to use the original crystal

GordonMcComb

The '256 is a fun chip to play around with, but it really wants to have that screwball crystal. Kenneth at Speechchips sometimes has these for a reasonable price. Worth it, to avoid the chip locking up.

If you're good at sounding out phonemes the SpeakJet and Soundgin aren't bad substitutes. Both are currently available, though the Soundgin has now been rebranded as the Babblebot. They were designed by the same person, using PICs as the underlying MCU. Speech quality is close to the SPO256, though it all depends on how they're programmed. Nothing beat the old SC-01, though. These are hoarded by HERO and RB5X collectors, and virtually impossible to find.

(Curiously enough, the SPO256 is believed to have been an early MCU, developed by General Instrument, who later sold portions of its business to Microchip. The SPO256 incorporates some analog circuitry not usually included in MCUs, so it's not a pure microcontroller, but it's interesting how things come full circle.)

Nice offer, and no, you shouldn't have too much trouble finding takers.

-- Gordon

PS: If you're looking for fun analog sound chips, it's hard to beat this one, used in all manner of toys and vintage acrade games, including Space Invaders:

http://www.bgmicro.com/ics76477.aspx

Good price, too. Not much more than what Radio Shack sold them for in the 80s.

erco

Gordon: As usual, you really know your stuff! I'm getting into speech big time now, I just bought two SC-01 chips, $25 each from Fred Teer, per http://www.redcedar.com/sc01.htm Per his email, he has plenty remaining:


From: fred teer [mailto:fredteer@yahoo.com]

Yes, I have quiet a few left. The price is $24.95 per device and $2.60 shipping.
For payment, I can accept PayPal or a personal check if you are willing to wait for the check to clear before I ship the product. If you like, I can provide you with my
home address and phone number. I do not have a business or a web site, these are items I collected almost 30 years ago.

CHEERS

Fred

RobotWorkshop

Erco,

I would recommend using the SpeakJet as a current speech chip. It works extremely well and reminds me a lot of the SC-01. Back in the December 2007 issue of SERVO magazine I had an article on how to use the SpeakJet as a replacement for the SC-01. The prototype would plug right into the SC-01 socket and emulate the chip so the device wouldn't know it wasn't talking to a real SC-01 chip. When the last of the SC-01 chips dry up then a small hybrid module can be made as a replacement.

To make the SpeakJet easier to use you can add the companion TTS256 chip that handles the Text To Speech for you. That is what I am running in the trash can robot I had at the last UPENE show. It also had a workout this weekend helping the Cub Scouts with their yearly popcorn sale..

If you want a higher end option there is always the V8600A module from RC Systems. They have been around for a while and their module sounds great and is easy to control. This is another one that I highly recommend. It is the voice for another one of my robots.

http://www.rcsys.com/modules.htm

Sadly many other promising options have just dried up. Gone are the EMIC modules based on the WinBond chips.....

Robert

erco

@Robert: I'll try Speakjet, too. Wanna see what works best. Any links to your trash can robot? Sounds interesting. Remember Jerry Rebman's GARCAN robot from the old TAB book on "advanced robotics"? An all-RC galvanized trash can robot, go figure.

@Gordon & blittled: Thanks for the heads-up on the 3.12 mhz crystal. Speech Chips is out of stock, so I just ordered 10x 3.2 mhz xtals off ebay ($5 total, incl ship). Hopefully that will get me going, and I can include a xtal with each 256s if I sell any.

erco

BTW, here's a guy selling a 256 with the matching text to speech chip for $55: http://www.ebay.com/itm/Radio-Shack-Text-Speech-Voice-Synthesizer-IC-chips-CTS256-AL2-SP0256-AL2-/220867357323?pt=LH_DefaultDomain_0&hash=item336cb86a8b

I bought a text to speech chip from the same seller previously for $20.

RobotWorkshop

erco wrote: »

@Robert: I'll try Speakjet, too. Wanna see what works best. Any links to your trash can robot? Sounds interesting. Remember Jerry Rebman's GARCAN robot from the old TAB book on "advanced robotics"? An all-RC galvanized trash can robot, go figure.

Sure. The outside looks similar to the Baxter robot by Steve Norris that he has in Robot magazine. It was built mainly with leftover parts and your favorite $10 gear motors that have the wheel already attached. I have my own H-Bridge board controlling them. It has two Propeller boards installed to control it. The top has a Ping module to scan around along with some extra LED lighting and an IR sensor for Sony remote. The Speech is done using the SpeakJet with TTS chip. I'm getting ready to add an extra internal sensor so it can tell if it is carrying something and also a few other external sensors.

You can see a picture here:

https://picasaweb.google.com/101575603280102512469/UPENE2011August272011#5645946959192693778

This one shows one of the spare motors w/encoder attached:

https://picasaweb.google.com/101575603280102512469/UPENE2011August272011#5645951512729351538

I would have had more at the show but there was only so much room in the car....

Robert

ctwardell

I've used 3.579Mhz with no problem other than I a little high in pitch, so the 3.20Mhz should do fine.

C.W.

GordonMcComb

Oddly enough (or not) crystals close to but not exactly 3.12 MHz may actually be worse than a timebase 300-600 kHz away, like a color burst crystal. I'm not sure why -- perhaps it has to do with baud rate setting and jitter -- but those close-but-no-cigar crystals may or may not work for you. You'll know you're circuit bent when you're sending the chip serial data and it just locks up on a phoneme - aahhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh. That sort of thing.

Good find on the SC-01's from Fred. Looks like his hoarding paid off! If/when his set is depleted, you might try John Boisvert (his friends call him Lord Hotwing), who now owns all the RB5X inventory -- that bot used an SC-01A in its speech board. I'm pretty sure Robert and John know each other, from eBay bidding wars if nothing else!

-- Gordon

erco

XLNT. I've read blogs from "Lord Hotwing" and his monthly meetings, he's somewhere out here near LA, I believe. A pretty cool gathering of robot fans/beer drinkers from what I gather.

The perfect marriage of the two interests, IMHO.

denno

For a long time...a long time ago, I did alot of work with the above. Then Parallax came out with the voice chip called the EMIC, which was, I believe built by Grand Idea Studios. I have several of these in operation, and they are great. Just load your text strings as DATA statements and access the address, and away he goes....blabing away. I don't see them on Parallax's product list anymore, but try Grand Idea's web site. The spo256 took alot of time to "teach" it to say the word the right way. If you are still interested in using the above, and want a bunch of words that I put together using the chips alliphones, I can send them to you. But the EMIC, if available, still is so much nicer for talking apps.

DennO

erco

Speech synthesis gets complex pretty quick. Glottal stops, voiceless fricatives, vocal tracts and more get modeled in this paper; have a peek at the "math of speech" at http://americanhistory.si.edu/archives/speechsynthesis/im_2469.htm

Humanoido

erco wrote: »

I can't wait any longer for a Prop-based voice synthesizer ...

Wait no longer! The Parallax Propeller is a speech synthesizer, it's available now, and it's a very good replacement for the SPO256. It can also do speech recognition. The software is free. Check these sources.

Page 16 post 305 Creating Brain Speech
* Parallax Propeller Takes the Lead as Speech Synthesizer Chip
* Introduction to Propeller Speech
* Development of Propeller Speech & Code Resources
http://forums.parallax.com/showthread.php?124495-Fill-the-Big-Brain&p=977506&viewfull=1#post977506

Page 16 post 303 Brain Speech Recognition
*Talk to the Brain in your own language
http://forums.parallax.com/showthread.php?124495-Fill-the-Big-Brain&p=976721&viewfull=1#post976721

erco

Thanks for getting all that info in one place. I have seen much of it in individual posts. Seems like there are several large significant chunks available, but no one has put all the pieces into one cohesive chunk. But people are interested. Per http://www.redcedar.com/sc01.htm :

"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."

Phil Pilgrim (PhiPi)

Chip's vocal track simulator has been available in the OBEX for a few years now. My phonemic synthesis program uses it, albeit poorly. I'm certain that his object has the chops to produce intelligible speech, but the devil is in the details; and those details have been -- so far -- a challenge to master. I think the solution to producing good speech is not to approach the task a priori, as I did, with assumptions about how each phoneme should be produced, but adaptively. In the latter approach, a program would be given an utterance to imitate, and then infer the best set of synth parameters to produce it. From there, the result could be dissected into its phonemic constituents for later recombination into different utterances.

In any adaptive system, one first has to develop a goodness-of-fit measure (utility function), i.e. a way to determine how much one simulated utterance is better than another. It may be tempting simply to perform a sum-of-squared-errors measure between two waveforms in the time domain or between two spectra in the frequency domain. But both methods have their flaws. For one thing, two entirely different waveforms may sound identical to the human ear; so eliminating a candidate set of simulation parameters, just because it doesn't match the sample simplistically, would be throwing out the baby with the bathwater.

Once a utility function has been arrived at, the next problem is how to represent candidate parameter sets in a way that they can be tweaked effectively to arrive at an optimum solution. Ideally, small parameter changes will result in small changes in the sound they produce.

Finally, a means for maximizing (or minimizing) the utility (or error) function, in order to steer the parameters to producing the best possible simulated utterance must be determined. This can range from simple gradient-following, to simulated annealing, to genetic algorithms. Of course, the resulting representation space has to map one-to-one to the parameters used by the synth object.

It's an utterly fascinating problem and one I'd be keen to tackle again, but it's something which requires more than just spare time to solve, I'm afraid.

-Phil

erco

Phil Pilgrim (PhiPi) wrote: »

It's an utterly fascinating problem and one I'd be keen to tackle again, but it's something which requires more than just spare time to solve, I'm afraid.

Eloquently worded as usual, PhiPi! You and others have definitely made significant strides on this. What frustrates me is that the 1980's "state of the art" chips (spo256 & votrax) did text-to-speech with some 300 linguistic rules in hardware, and it seems that would be relatively easy to duplicate in software. Is it an IP or licensing issue which prevents using the existing 30-year-old technology? Or is it just a matter of Parallax or someone else making it a priority and coming up with sufficient cash to take it off the back burner? Per the limitations of "open source" design, they would have to give the software away free, but that would certainly drive Propeller sales.

shimniok

Glad to have stumbled across the thread. I bought one of the last 256's on the shelves of my local RS in the late 80's. Hooked it up to the ol' C64 using a 3.579Mhz crystal (that I still have) and it always worked more or less fine. Not long ago I got it working again (it's still on the same breadboard I put it on 20 years ago) using an ATmega328 / Arduino. Would love to put the Propeller to use with this old chip (or doing speech in software). Fascinating topic... time to read some links above...

Humanoido

When I wrote a simple talking Propeller keyboard program, I discovered two ways to induce speech. One is character by character and the other is by sound. If you input and concatenate the sounds in a raw fashion, it will do a nice text to speech. There are no extended sets of rules so this is just a simple demonstration, or, if one learns the sounds, it could be much more understandable. For reciting known texts, the object is then to go back and put in your own changes to adjust the sound. Essentially one is making the rules for sound adjustment.

erco

Fred Teer is selling his Votrax SC-01's on EBay now, FYI : http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=230672354470

erco

Per Gordon, here's Babblebot, a $20 voice/sound chip contender : http://www.speechchips.com/shop/item.aspx?itemid=21

I haven't heard any speech samples yet.

Leon

It looks like a PIC, from the pin outs. I'm surprised that they didn't use a dsPIC.

I found this:

http://babblebot.net/babblebot_ic.html

which confirms that it is a PIC. With a bit more digging I found that it's a PIC18F1320:

http://www.uelectronics.info/soundgin-a-programmable-sound-generator

erco

Found one dreadful sound sample at http://babblebot.net/index.html

Sounds like "daddledot world dot" ... Surely that can't be their best?

But I saw no other samples on their home page. Might be a bad sign.

Leon

Here is an example of the output:

http://www.ginsingsound.com/

It isn't bad, given the limitations of the chip they are using. A dsPIC can produce good quality speech, using the free Microchip Speex library. Here are some samples:

http://www.microchip.com/stellent/idcplg?IdcService=SS_GET_PAGE&nodeId=1406&dDocName=en023610

It isn't really synthesis, though.

I found this:

http://home.alphalink.com.au/~derekw/pictalker/main.htm

which uses an old low-end PIC. It might be possible to filch his allophones and use them with a Propeller.