• Please review our updated Terms and Rules here

the newbie's collection

patscc

Veteran Member
Joined
Apr 13, 2005
Messages
2,311
Location
Stevens Point, Wisconsin
Okay, being the newbie, guess I should post my list.

NeXT cube, with laser printer.

Epson QX-10 w. Titan 8088 daughterboard
Epson PX-8, just added a second one
Epson HX-20, two, with speech synthesizer

TRS80 model 1 level 2
TRS80 model 4
TRS80 model 4D
several TRS80 model 4P's


various Sharp PC1500's, Tandy PC1's, PC'4's, I think I have a PC7 around somewhere, maybe a PC2, maybe a PC1600 (I'm moving, hence I can't remember without unpacking)
a couple of Sharp PC2500's
Sharp PC5000 (PC clone, uses bubble-memory carts)


a couple Tandy Model 100's
Tandy model 102
a couple Tandy Model 200's
a couple Tandy Model 600's, one with cracked display

NEC PC8201A
a couple NEC PC3000's


Tandy1000EX
Tandy1000HX( at least, I think it's the HX, it's packed)

a couple Tandy 1100FD
Tandy 1100HD
Tandy 1400LT
Tandy 1500HD

Amstrad Portable PPC640, which has got to be the wierdest-looking PC clone I have. It runs of D-cells, and has a full size AT-style keyboard, if you can picture that, and a 7" foldout LCD screen.

several Toshiba T1000's
Toshiba 1100plus
a couple Toshiba T1200XE
Toshiba 1600
more, somewhere, can't find, can't remember

TI VDU200, some luggable with a TMS9900 -based "cage" inside it, still trying to figure this one out.

TI994A
a few TI CC40's with RS323C cart. I love these. Only TI would market something with a great big sticker on the front "The WaferTape drive is not available"


HP85
HP15C
HP41GX

a couple HP110, one has bad memory, it's a GetToIt
HP110+

various Compaq luggables, from big to small
IBM Portable, the one with the Amber screen
a couple of IBM convertibles, one with backlit display, one without( NOT broken, appearantly they were actually different)
IBM P70

of course, IBM PC, XT, AT

a couple of Mattel Aquarius
Coleco Adam

Kaypro 2
Kapyro 2X
Osborne 1
Otrona Executive
a couple Commodore SX-64's

a couple Sinclair ZX81's
Timex/Sinclair 1500

Atari Mega ST
Atari 600XL
Atari 400

Zenith 180's

bunch of old Macs and Mac-junk

No S100 stuff yet, sob. I'll get there.
Oddly enough, no Apple II's. Can't seem to find one around here.

386 and later are mostly clones.

all this sits on a big raft lashed together from C64's, in case the basement floods. There's probably some VIC20's in there as well, and a Plus4

Just re-found CBM 4000 series dual drive unit, but can't seem to place it, nor can I remember if I have the actual computer that goes along with it

Probably missing some stuff, but a good portion of it is packed up for the big move, well, I lie, I've still got a tone of stuff and docs to pack.

patscc
 
Very impressive collection. Do C=64s really float? I know too well the flooding-basement situation, I had that problem once, after living in that house for 8 years (just long enough to catch me with my guard down). Took me about a month to get everything rescued (cleaned up & dried out). I had over 150 computers at that time, one of the reasons I decided to down-size my collection to something more managable. I like to keep it somewhere around 50 machines these days.

--T
 
Well, there's more stuff, like CoCo's and uCoco, various luggables, one of the old Zenith lunchboxes that boot to a world map, and so on, but I got depressed thinking of all the effort it'll be to pack it all up, so I stopped the list.
Wish I had some S100 stuff, though.


Well, I dug out a particularily grimy C, filled the tub, and it tilted vertically, and looked like it was settiling with the narrow front pointed up. So I ran downstairs to get a camera, and when I got back upstairs, it had of course sunk, much like the UPS you never get around to replacing the batteries in.
Thank God my wife is out running errands. We have completely different ideas as to what tubs should be used for.

I think if I duct-tape the cooling vents, I might actually get it to float long enough to take a picture. If I can, I'll post a pic somewhere.
Anyways, this means my COOP-plan requires rethinking. Maybe I should switch to Sinclairs ? There's gotta be something I have that'll float. Gotta make sure the next place is not on a flood plain, though.

Seriuosly, what I try to do is keep everything in one piece up the first foot or so on shelves, and use the lower ones for boards and stuff in pieces,in hopes that it will at least be easier to clean should it flood.
Of course, what's probably going to happen is as soon as I have everything in storage, the storage unit's roof will start to leak.

patscc
 
I dunno 'bout floating, but I have the next-best thing. My Itronix handheld is supposed to continue functioning even if submerged under a foot of water. (Of course, I have never tested the company's claim, but I have given it the "drop-test" several time).

--T
 
That's actually quite usefull. That means you can sit in your flooding basement and redo your calcuations on how quickly it would take to bucket it out now that the sump's shorted out, even with water closing over your head.
Of course, now I'm totally fascinated with this idea. I guess something that doesn't have a lot of expansion ports, and preferrably a rubber keyboard, like maybe a CC40, or a Aquarius ?
patscc
 
BTW, are your HX-20 speech units the Epson RealVoice? I've always beem fascinated with them. How do they work? Is there tape-based software for them, or is it on a ROM? Where is the ROM located, in the HX-20's internal socket, or do you need the ROM expansion unit, or does the ROM mount inside the speech unit itself? Is the software for it written in BASIC, or Machine Code? Does it really do it's text-to-speech majik in only 16K, or does the speech unit contain extra RAM? How large is it's built-in vocabulary? Does it allow the user to add new words? Does it have pre-programmed phonemes to create new words? Oh yeah, are yours male or female?

--T
 
They are indeed. The one that's still out is Male. They're still a fairly new( I got them about 3 months ago ) addition to the basement, and work's kept me busy so all I've been able to do is pull it apart.

The main chip has had the "sandpaper" treatment, and by the usual method of licking it and angling it in a flourescent light, I was able to make out HD637B01, so I think it's out of the Hitachi HD63701 microcontroller family, of which I've no spec. It's a 64 pin DIP, but not the .100 spacing, it's the next-smaller one down, so it looks, size wise, like any ol' 40-pin chip.

The software is contained on one 27C512 eprom. Additionally, there's a 32k scratch-pad SRAM for the microcontroller.

I think it's entirely phonem based. The voice board contains 11 27C512's (up to 12) for a total of 704 kByte of phenom-space. This, for it's day, is not bad. It is also oddly enough more memory than DOS knew what to do with for a long time.

I've also worked with early versions of Microsoft's speech SDK, up to 5.0 (2001) and aside from having a truly massive footprint, doesn't sound much better than the RealVoice.

The way it works is thus:

You attach it to the HX20 by way of the expansion port. That's all. No cassettes, no ROM-packs.
Power on the HX20.
It says "Hello"
You type what you want, e.g. "I am a blue smurf" and hit enter.
It speaks. Simple.

It also has a remote ear jacks aside from the internal speaker.

Being phonem based, you can't add words, but then you don't need to. I mean, if it can say 'Smurf', what more could you want ?

The one I have is mated to the HX20 by extra reinforcement bars attached on the underside between the two units, to keep them stable. It came from a mobility cart for disabled foks, and whas bolted to the frame, hence the "ruggadizing". I love stuff that's beat up and modded, hinting at a long and (there's that word again) useful life. The other HX20 has had a hole drilled in the side, and an external RC NiCad pack wired in, and taped to the underside. Maybe some poor person's grandkid that had a knack for fixing stuff modded it, since getting the original NiCad pack probably cost an arm or a leg.
That's the great thing about vintage computers, when they have personality.

Oh, I have a recording of the RealVoice synth, so if some one wants to here it, pm me or give me a place to post.

patscc
 
I'm looking at my 1985 Hitachi databook, and there's two different versions listed, HD637B01V0 & HD637B01X0, the difference being in the number of i/o ports, and of course, the packaging. The V0 is a 40-pin ceramic DIP with a program/erase window (both types contain 4K of ROM & 192bytes of RAM). The X0 is the 64-pin chip with 53 i/o pins. It comes in two different packages, both 64-pin DIPs. The ceramic package has the EPROM window, but the plastic pack is one-time only programmable. Both units are listed in my book as -preliminary-, so they aren't full-blown datasheets, but if you'd like, I could dig out my scanner and send you copies. The block diagrams might be helpful.
The RealVoice not only ain't bad for it's day, it's still pretty good for any day. Many of them are still on the job. One of the BBSs that I used to frequent was the MATCHBBS (Michigan Assistive Technology Clearing House), and quite a few of the deaf population in Michigan still use them, or did up until Y2K, when the board had to shut down because the Galacticom (WorldPort) software they were running on went AWOL, and I lost touch. (I lost my (last) free internet access at the same time, and had to start paying an ISP for the first time since 1992). That's one reason they fetch such high prices when they sell, because there's still a big demand for them within the deaf community. In fact, I almost wouldn't have the heart to collect one, knowing that somone with a disability needs it more than I do.
So, with the RealVoice mounted, the HX-20 becomes a dedicated talking machine? You can no longer program it in BASIC? Or does it just add an entry into the Menu for RealVoice?

--T
 
It's gotta be the XO, no window. Yeah, scans would be helpfull, no rush.
It replaces the usual
1. Monitor
2. Basic
menu you get after a cold reset.

You get a dedicated talking machine.

Your point about their usefulness is well-taken. I'll keep my eye out for them, and if I encounter more of them, I'll scout around for a deaf/mute user that needs one, and sell at cost. It's great that they're still usefull, and it really sucks that anyone should have to pay any sort of premium for access to something like this. Silly me, I thought all this stuff had been suplemented by newer stuff.

You can actually operate everything blind. It's really amazing. After your post it struck me, and I pulled a thick woolen hat over my head and turned off the lights. Here's what I noticed.

The dogs were not happy with this.
I can feel and find the on/off slide for the HX20 in the dark.
The keyboard has a good 'click' feel to it, so I can tell if a key has been fully engaged.
It's easy to tell where one key begins and ends by feel if you're not used to typing.
The HX20 is robust. There's nothing moving about in there, if I bump into it and it drops, it still works. ( I just now discovered this the hard way)

Oh, and having recently had to suffer through Verizon's "tech" support, I have come to the conclusion that it takes a whole lot less than 704k to simulate a human being, and only takes a couple of bits to truly make you feel like a monkey.

It's truly an amazing setup, the more I play around with it.
By the way, does any one remember Ray Kurzweil ?
patscc
 
Y'know, the more I think about it, the more likely it seems that RealVoice must have a built-in vocabulary, and just uses the phonemes for words that ain't in the dictionary. How else do you account for it having >700K of ROM? The English language only has about 60-sum'n phonemes, and a couple K should be plenty to store 'em all, but 700K would certainly store a lot of complete words. Of course, that's a lot of RAM too, I wonder if it does allow you to store new words somehow? (I'm guessing you don't have a tech manual).

--T
 
Yup. No tech manual. Voice is tricky. Here goes...

By the way, off and one, I'vve been replaying Wizrardy8, and so currently I'm getting pherome and phonem mixed up. I mean phonem, in all tha tfollows.

Y'all might want to skip this, this is one of my pet fields( what I get paid for) so I do tend to go on.

You're right, there's a limited number of phonemes, somewhat language dependant.

It's actually about 43( English , more, if you're French, and after all, they do need some kind of excuse for their beer) if you talk to a linguist, or *a whole lot more* if you talk to a programmer, because, reagrdless of what linguists say, we treat( and tend to digitize) as many diphtongs as we can ( i.e. as computer memory permits ) along with the basic phonems. Also, it depends on how many beers we've had, since the slur density per coherent words occasionally makes elimination of noise or redundancy possible along the lines of "Yo, don't be sick on my bar..., don't, aww, get him out'ta here"

Non-linear voice compression, which is what this is, dates back to when MaBell diecovered it's cheaper to make chips than relays, and cheaper to have a computer route calls than an operator.

MaBell, at some point, decided to define 300~3000 Hz as the majority spectrum of human voice. Bear on...

The basic unit in telco land is a T1, which spins along at around 1.5 Mbps.
This was done decades ago. Oddly enough, T1 is a serial inerface.
This is subdivided int 24 timeslots, each is about 64kbs wide.
It was probably more like 24 is a nice number, and 64, well, all geeks like 64, put the two together and we've got 1.536 Mbps. Uh, well, 64 and 24 are still nice.
The 24 came about from long experience, remember. each residence gets their own wire pair, they go to a pole, are gathered in a trunk...
24 pairs work well, bundle them together, the *thick* cable is managable, and it worked well with MaBell's 1st gen of trunk multiplexing, which basically went thus.
Okay, fiirst channel, goes through under 4000 Hz.
Instal heavy-duty lowpass where channel enters telco.
Next channel, mix onto carrier frequency to put lowest range above 4khz.
.
.
Channel X, mix onto carrier so as not to...
.
.
Oddly enough, DSL works much the same way. It's called, I tink, freqeuncy multiplexing, or who knows.

This mux process was very sensitive to distance, and so on.
Sound familar ?(I'm sorry, you're too far from the switch...)

Back to T1.
This 64kbps channel is one of the reason's modems top out at 56kbps. There's more, a lot more technical reasons, but this'll do.

So, for practical purposes(Nydquist +err) , for telco quality voice, you need to sample at at least 6100 samples per sec..

At some point, some one decided it would be handy to reduce this bandwidth for storage purposes.

There are quite a few schemes to reduce this rate. Telco doesn't care, really, since 1 channel = 1 line = 1 voice equivalent circuit is set in stone, and is unlikely to change as long as engineers that understand this stuff are still alive( When they're all dead, then, HAHAHA, I can take over the world with my evil plot ! I'll call it T1a. The details will follow in the same theatre...)
The people who care are the ones that have to store this stuff.
Back to 'Speak and Spell'
Can't find my specs, but the effective compression was incredible to be able to stick it in a cheap ROM.

A lot of this stuff had to deal with how stuff is digitized, involves calculus and discrete math, and since I don't want to be beaten, either becuase it's boring, or I'm screwing it up since I've already packed up my NeXt which was the only system I had left running Mathematica) I'll skip it unless someone offers me a job.

This is where phonems come in. Phonems were viewed as a "short-cut" to the common fragment you could gleam from vocoder theory.

Phonems are the 'common' phonetic, but really, the basic set of grunts, fragments of a particular language, i.e. the belches. Ever notice how you can always tell when someone's belched ? Even though the particulars might vary ?

This is what phonem breakdown does. It avoids the mathametically complex breakdown of the vocoder( i.e. leaving out the nerds ) and comes up with a shortcut of real-voice fragments that are close to what the mathamatician was probably talking about )
Sorry I left out the engineer, but funding hasn't actually been approved yet, so why bother :)

*whew*

So, anyway, take a raw sample rate of around 6.2k, and we divide that into 704k that the HX20 RealVoice uses, and we end up with round 113 seconds, enough for some high quality phonemes, the programmer's realistic approach of including dipthongs, and the marketing depatrment's "The next scheduled service is due in ..."

Remember, in the classic phonem, ther's no room for inflection, and anybody that's ever had to implement this realizes you have to be able to generate some sort of inflection if you don't want your system to be treated as a taxi cab. (Brrrr... bad word in Joisey, taxi )


This makes me think the phenome encoding is 8 bit, around 8 khz sample rate in the HX20 realVoice.

Damn you, Terry Yager. Now I wil have to unpack this device, and quest for the D/A chip in it, because now I need to know what the conversion width is. (The telco we all love uses 12-bits. Don't ask why. Maybe they decided 16-bit was decadent, but they could slap 3 16-bit shift registers together, and fake 4 12-bit channels, I don't know. Neither did they. Look what happend to the Video Phone) Actually, it's worse that that, for various reaons, for signalling purposes we decided to trash every other LSB in the stream. This is way off topic, PM if you want particulars.

No, brainstorm, just hit me. Telco couldn't do 8 bit conversion because of the noise floor presented by good ol' telco residential wiring, and keep up the loss assumed by robbed-bit siganlling.

Probably, to date, the most incredible job of speech compression ever done with english was with the TI 'Speak and Spell' toy. Sure, it really sounded choppy. It hardly took any space at all, ran off a calculator chip, and, being a phonem based system, could technically synthesize any possible english word. (Of course, TI being TI, they used a chip to actually synthesize the phonems that they'd derived from doing 10th order LPC( linear predictive coding), one of the reasons it sounds so choppy)

It used a (at least) 10th order LPC process to generate the phonem fragments. Incidentlly, something similar to this is what military comm systems use today as fail-back when regular clear voice systems break down. They break down incoming voice into LPC codecs, transmit the codec tag reduntantly( think Reid-Solomonn, but of course, what else is there, really) recombine, plop out the phonem, and bingo, voice. Oh, and for the benefit of Messrs. Raptor and Carnivore, and their progeny, military comm channel redundancy and bandwidth estimates are easily accessible via google. And froogle. And spam. Why do I get spam advertising stuff I'm not supposed to know about ? Am I missing something ?
Seeing as how I work in DC, I'll charge military and fed gov the DISCOUNTED (KICKBACK $$$) rate of 1144.66 per day for tracking down google references; I fully appreciate the diffuculties on the proper usage of Google in an agency environmet.

Although, strictly from a taxpayers perspective, it is interesting to note that the TI SpeakAndSpell went for, what, $25.00, and the reduced bandwidth military comm systems go for, well...
Think about this way. How many Speak and Spells were sold ??

10,000
100,000
1,000,000

Can you see the military bringin it in for less than 25,000,000 ? And that's assmuing 1,000,000 units where implemented, and not, like, 22, or 30, of 43 ?

Aaahh, I need to learn to type slower, or something.
This is where the <snip> command comes in handy.

So, back to Terry's question, once you've implemented a functioning phonem set, you don't need new word storage.

What's so nice about this, and this goes back to a pet peeve of mine regarding character sets, is you type in your sentence, cheap, not much storage, then you do something like the following:

Of course, you've got an index set to favour the organization of your phenoms.


Somewhere along the line, you've determined a 'dummy count"
You're scanning incoming typed characters for garbage. (trivial, I think there's actually mention of this in the original Kaypro bios code)
So, sit on dummy count, you hit it look for match, miss, abort, flag user,
No dummy, keep typing...

The truly brilliant part lays in breaking the stream into phonems.
Terry mentioned 'is it all done in 16k', referring, I think to the base memory of a HX20, and I think that's about right give or take some scratchpad.
Remeber, this is all done on a notebook that is from the geriatric era of 8-bit machines.

One of these days I'll actually manage to crank out a terse post.

patscc
 
OWWWwww! My head hurts! Seriously tho, I think I did manage to get my brain wrapped around a small part of what you said, I just need a little time to digest it.

--T
 
Back
Top