EEVblog #713 - VCP200 Voice Recognition - 1980's Style

[EEVblog]

Dave breadboards a 1988 vintage Tandy / Radio Shack VCP200 speaker independent voice recognition chip from Voice Control Products Inc.
Will it work after 27 years?
Will it recognise his Australian accent?

Radio Electronics magazine April 1991: http://ia601702.us.archive.org/21/items/radio_electronics_1991-04/Radio_Electronics_April_1991.pdf

MC68HC04 Datasheet:
http://images.ihscontent.net/vipimages/VipMasterIC/IC/MOTO/MOTOD005/MOTOD005-3-355.pdf

M6804 User Manual: http://www.textfiles.com/bitsavers/pdf/motorola/6804/M6804_MCU_Manual_Sep85.pdf

[krivx]

I was wondering how a product with such limited market (hobbyists mostly?) could cover the cost of spinning silicon - a maskrom mcu makes total sense, maybe it was even financially successful.

[sleemanj]

Brings back memories of a voice recognition expansion cartridge for the C64 a way back when.  Invariably, it devolved into my shouting at it in various different intonations in order to try and get it to recognise words.

"Hello computer"
"Hhheeelllloooo coooommmmppppuuuttteeerrrr"
"Hello COMputer"
"HeLLO comPUTer"
"HELLO COMPUTER"
"GODDAMIT HELLO COMPUTER"

[pickle9000]

You may have a new series, Electronics for 3 year olds.

The circuit breakdown was very well done. A beginner with a scope should find it very useful.
 

[live1052]

That kid of yours is unreal! Hard to believe he's only 3. Great to see he's taking an interest. I'd hope any kids I have would at least have a passing interest in my hobbies!

Dave, I think you should break out the Tandy 200 in 1 soon and do something creative with what's on the board in honour of Radio Shack's closure!

[benst]

Love Sagan there. I don't know why you bother getting an intern really. Seems to me Sagan is better at doing electronics than 80% of the interns I've had.

[JackOfVA]

As a guess, the firmware uses a Goertzel-based tone detector algorithm.  This is much more efficient than a classical FFT when run on a small micro controller.  Typically one can achieve 25 to 50 Hz wide bins with the Goertzel algorithm, but the bin width and number of bins calculated per second is a function of the micro controller CPU power.

Based on the power / voltage per bin, the software can make an intelligent guess at the command being spoken.

A few years ago, I built and marketed a Goertzel-based tuning indicator for Morse code and radio teletype tuning. http://www.cliftonlaboratories.com/Documents/Z100%20Tuner%20Operating%20Manual.pdf for the assembly manual with some theory of operation.

[jmc734]

Forrest Mims published an interesting article on the VCP200 in the January 1989 issue of Modern Electronics (http://www.americanradiohistory.com/Archive-Modern-Electronics/Modern-Electronics-1989-01.pdf)

[dentaku]

That works surprisingly well. At least it wasn't $169.95 like the TRS80 Vox Box.
Of course it had the immense power of the TRS 80 to process speech
http://www.trs-80.org/trs-80-voxboxhttp://www.trs-80.org/trs-80-voxbox/

[Paul Moir]

Real minor, but the yellowing of the packaging has nothing to do with bromides.  It's just something that naturally happens with transparent PVC.  I know high energy UV will do it quickly, but I think it might happen in enough time too.  Also, there's very little reason to put fire retardants in packaging materials.

I couldn't afford those VC2000s when they were available.  Glad to see I wasn't missing much!  I do wonder though if a schmitt trigger on the input might have helped?

At 17:06, is Sagan putting double quotes around "side cutters" with his hands? 

[NiHaoMike]

What would be interesting is to see how much better could be done with a modern microcontroller in the same or lower pin count.

[Agent24]

  I do wonder though if a schmitt trigger on the input might have helped?

That's what I was thinking too... maybe it didn't work, or maybe they didn't want to add another IC?

[AndersO]

I remember drooling over this exact item in the RadioShack catalog or store when I was about 10. I didn't have the skills and never would have been able to build and test the circuit at the time. Still might not without some trouble. So this was so rewarding to see. It is almost magic to see it work still today.

Seems really similar to the VRC008 which is apparently also a 6804 based voice recognition chip made by Interstate Corp. The way the software works is described in US patent 4388495.
Link: http://www.google.com/patents/US4388495

"The microcomputer, by monitoring zero crossings, defines words as a sequence of vowel-like, fricative-like, and silence states. By limiting the recognizable vocabulary to words which have dissimilar sequences, the incoming speech pattern may be recognized by comparison with state templates defining the limited vocabulary stored in the microcomputer's memory."

As you guessed it has the element of simple "state machine". I'm guessing this is either a derivative or a very similar design. Explains why you need Turn Right v.s. Left Turn. You can find datasheets scans for the VRC008 from all the usual questionable datasheet online places. Very similar, so it's a good guess thats how they did it too.

[richfiles]

Radio Shack may be going the way of the dodo soon... But color me surprised when I found they STILL have a support page for this old chip! 
http://support.radioshack.com/support_supplies/15365.htm

[radiomog]

I have an old RS voice recorder chip I should do the same thing with.. same vintage/packaging as your chip.
I found it going through some old bins of parts in my lab.

[Hade]

FYI the forum link on the blog post is broken.

[Fungus]

I remember seeing those in the local Tandy! 

[max_torque]

Dave, if you've still got the circuit on the breadboard, i'd love to see some screen captures from the scope of the digital edges from "ON" and "OFF" or similar, just to see what the difference in the digital signal actually looks like!  (these days with a Storage scope, that's easy to do,  in 1988, it would have been much more difficult (or expensive!))

[Quarlo Klobrigney]

My thought exactly. Dave if you still have it breadboarded, how about a 74HC14 off the output of the comparator coupled somehow? That should square things off.

[G7PSK]

I have one of those VCP chips from Tandy 1988 vintage. I got it in a big bag of parts when the Tandy store in Kings Lynn Norfolk closed down. I had been thinking of sending it to Dave but never got around to digging it out, I now have so assuming Dave does not want it is there any one who does it is of no real interest to me and as far as I can tell not a lot of practical use either, it is still in a rather yellow and brittle plastic case along with its documents. I am not looking for payment but if you want to send Sagan something I am sure that would be appreciated.

[jancumps]

That kid of yours is unreal! Hard to believe he's only 3. Great to see he's taking an interest. I'd hope any kids I have would at least have a passing interest in my hobbies!

Dave, I think you should break out the Tandy 200 in 1 soon and do something creative with what's on the board in honour of Radio Shack's closure!

I've made the reverse circuit with my 200-in-1 and a microcontroller a few weeks ago. Make it speak phonemes

[mux]

Real minor, but the yellowing of the packaging has nothing to do with bromides.  It's just something that naturally happens with transparent PVC.  I know high energy UV will do it quickly, but I think it might happen in enough time too.  Also, there's very little reason to put fire retardants in packaging materials.

I was wondering this when he said it - I can't remember bromide being added to PE and PVC packaging material. It might very well simply brown because the natural color of PVC is brownish yellow, like PA (which most people know as white nylon). Over time the additives that make it transparent may be unstable. Just conjecture, though.

[Quarlo Klobrigney]

I've made the reverse circuit with my 200-in-1 and a microcontroller a few weeks ago. Make it speak phonemes

The SPO256?

[orolo]

Great post, really instructive! Just a doubt: why isn't the signal into the comparator decoupled by a cap? I mean, jfet style, the audio enters by a cap, then a big resistor to ground for DC coupling. Wouldn't that, and a bit hysteresis in the positive side, help?
Cheers, and thank you for the great posts! I'm currently tearing down an intercom, something I would'nt have dreamt of without your great tutorials.

[jancumps]

I've made the reverse circuit with my 200-in-1 and a microcontroller a few weeks ago. Make it speak phonemes

The SPO256?

No, an Arduino.
Made it speak out the payload it receives from internet messages. The 200-in-1 used as a low pass filter to mill the corners off of the PWM output, and two of its transistors as amp.

[Jope]

This chip reminded me of an article I read some while back in Circuit Cellar about voice recognition with a 8-bit microcontroller.
Fortunately, you can download it as PDF: Low-Cost Voice Recognition, by Brad Stewart, 1998.
If you look at the schematic and figure 1, you will see some commonalities with the VCP200.

[Excavatoree]

I continue to be surprised by the people who complain about Dave including his son in his videos.

I've given up replying to their comments, but two thoughts come immediately to mind:
1.  No one is forcing them to watch.  (If they are being forced, then call the police.  Don't type a comment.)
2.  There is a feature that Youtube provides that allows one to scroll past bits one doesn't wish to see.

I personally like seeing Sagan - how he's grown, what he's learned.  Reminds me of myself and my friend, who's father was also an electrical engineer.  I'm amazed at how articulate he is at 3, and how he doesn't just parrot Dave.  (Someone mentioned in the comments that Sagan says 'ell-ee-dee"  not "led" as Dave does.   

I have that IC - I never did anything with it.  I think it was on clearance.  I'll have to find it.

[selkathguy]

At 15 minutes you mention that it is important to have the threshold value lower than the offset value of the gain stages.  This is counter-intuitive to me.   I would think that setup would compare most of a sine wave to the positive rail(?)

[gonzoid]

Don't have a single 10MHz crystal in the lab it seems. Sent the wife to @JaycarAU to get one :->

From https://twitter.com/eevblog/status/564999135640944641

Never heard of this "the wife" provider. Would you recommend their services as a reliable supply chain (package as requested, no counterfeit parts, etc.)?

[Towger]

Never heard of this "the wife" provider.

Must be a stop gap measure while he waits for Free_Electron to deliver the free Tesla...

[mtdoc]

Very nice Video Dave!

I especially appreciated the "fundamental Friday" type explanation of the front end circuit.

Nice little chip for its day.

I'm dismayed about the continued limitations of voice recognition technology - at least the software sold to professionals.  I'm frustrated everyday by the limitations and errors produced by Dragon Medical - supposedly the state of the art for the industry.

Oh well, back to dictating patient notes:      Headache, head - ache, HEADACHE, I said headache dammit....

[Mike Warren]

I'm dismayed about the continued limitations of voice recognition technology - at least the software sold to professionals. 

I've noticed the same thing, but Siri seems to work impressively well, from a short demo by an iPhone-owning workmate.

[Riotpack]

Could you change the bandpass to a 18db rolloff type and tighten the bandwidth to accommodate male only for your application. Maybe even a noise gate at the end?

[mtdoc]

I've noticed the same thing, but Siri seems to work impressively well, from a short demo by an iPhone-owning workmate.

Google voice is pretty good too, but they both rely on sending the recorded speech off to a server for processing.

My car has fairly good speech recognition as well. It's an EV but even so there is a fair bit of road noise in the UK due to our low quality road surface materials. It still manages to recognize commands and names in my phone book with about 95% accuracy.

Yes, for simple command type needs , speech recognition is pretty good these days. Unfortunately when it comes to more complex needs - like transcribing a complex narrative - the kind of thing a human transcriptionist can do well - the technology still sucks!

[Smokey]

The guy that wrote the article in 1991 was named D.B. Cooper!  He must have started an electronics company with his bag of cash.

[radiomog]

I have an old RS voice recorder chip I should do the same thing with.. same vintage/packaging as your chip.
I found it going through some old bins of parts in my lab.

finally remembered to take a photo....
not as old as Dave's, but pretty close

[Tek_TDS220]

Paul Moir is correct: the yellowing of the packaging (probably PVC) is not due to bromine.

I thoroughly enjoyed this video.  I remember seeing this IC in a local store, and wondering what application could tolerate the likely accuracy rate.  How high does it need to be useful - >99%?  Still, amazing result for the time and 1.6 KB program space.

I also enjoy seeing short clips of Sagan - he reminds me of my grandson.  Enjoy him while you can.  Time is a jet plane - it moves too fast.

[SeanB]

Blow moulded, so likely PVC film, which will yellow and go very brittle with time. I have a roll of PVC film that I stored for 10 years rolled up, and the outer layers and the edges are going brittle with age, just from being exposed to air. It is stored in a dark place, so it is not light affecting it, though light will hasten the embrittlement.

Just hope we use up all our film before it goes like that, but as this roll was already 5 years old when I got it 10 years ago as a test roll from a batch, the new batch of Swiss made stuff should last out till it is used up. Only about 2 tons left.

[Dave Turner]

Voice recognition and artificial intelligence - a dream for many years - particularly for those of us who watched the original Star Trek series in the 60's. There were, apparently, a number of research programmes trying to make the dream become reality. It's hardly surprising that an early barely functioning chip was produced bought.

For a few pence/cents one used to able to buy 'X-Ray Specs', clearly a 'con', nevertheless children still bought them. At least the VCP200 functioned to some degree.



 

[joeqsmith]

Thanks for the post!  I Really enjoyed this video.   

Many years ago, before the first sound blaster card, I was attempting to make my own speech recognition system.   I had seen speaker independent system at Western Electric/Bell Labs in the late 70's early 80s and became interested in it.   I still have my home made board laying around.   It may not work as good as the radio shack part.   

During my first job, we were making those LCP encoded speech modules for cars.  I still have some old audio tapes from back then as companies were trying to come up with more natural sounds. 

[joeqsmith]

Dusted off my home made recognition system which can be seen here: 

https://www.eevblog.com/forum/projects/retro-home-made-speech-recognition-system/

Again, thanks for posting the original Radio Shack video.

[zal42]

I know this is a moldy oldie topic now, but I just saw this episode and thought I could shed some light on this matter. I worked for a speech recognition company around that time period (not the same one), and I am 95% certain that they have implemented the same sort of tech were were using in the day.

It's much simpler than it sounds, gives workable results, and requires very little processing. Essentially, what it does is count the zero crossings in the speech recording and keep track of how much time has passed between them. It discards all other information.

The timings are threshold filtered and normalized (so that absolute speed of speech doesn't matter, only the ratios of the zero crossing times), and the resulting timing set is compared to a table. The one that it looks the most like is the winner.

I am, of course, glossing over fiddly details, but this is how it works. All the math required is simple and can be done with 8 bit integers.

This chip reminded me of an article I read some while back in Circuit Cellar about voice recognition with a 8-bit microcontroller.
Fortunately, you can download it as PDF: Low-Cost Voice Recognition, by Brad Stewart, 1998.
If you look at the schematic and figure 1, you will see some commonalities with the VCP200.

Yes, I worked with Brad Stewart (I still see him occasionally). That article describes an early incarnation of this method.

[EEVblog]

Thanks for the info.