Intellivision development, back in the day

[decle]

Hi everyone,

I think some of you may be aware that two academics from UCI Irvine, Tom and Braxton, are currently writing a book on the history of the Intellivision. What you might not be aware of is that JoeZ and myself have been helping them with a little technical consultancy here and there.

As part of their work Tom and Braxton have been granted access to the BSR document archive that was amassed by Keith Robinson. It turns out that this is a bit of a technical goldmine. The BSRs have kindly given permission for some of these docs to be reviewed and shared (many thanks guys).

So here is the first document of interest to this thread, the APh Datawidget Users Guide:

 

aphDatawidgetUsersManual.pdf

This describes the architecture and use of the Datawidget in general, and its Intellivision incarnation specifically. This confirms much of David Rolfe's overview as recorded in the development description and provides lots more details. As a result of this I have updated the development document:

 

intellivisionDevelopmentBackInTheDay-20190321.pdf

 

I should have some further updates as we get more details. As always any corrections or additions please let me know.


Cheers

decle

Edited March 21, 2019 by decle

[intvnut]

I'm always pleasantly surprised when bits we speculated on turn out to be correct, such as the model of terminal being used, or the fact that a T-Card provided the additional RAM at $5000 - $6FFF rather than a circuit modification on the Keyboard Component.

[decle]

Hey all,

Here we go, bit of a bigger update this time.

Within the BSR archive there are a number of technical bulletins covering the period from September 1982 through to October 1983. Once again, Tom and Braxton have kindly scanned these for us. The bulletins seem to have been memos sharing updates from the various infrastructure and support teams with the group as a whole. They cover many aspects of development, including tools for programming M-Network games for the Atari 2600 and Apple II; and also Aquarius work. The cool thing is that, because they're dated, we can put together a bit of a chronology. As a consequence, I've updated the development description again with a load of additional details and references to the specific bulletins that provide supporting evidence:

intellivisionDevelopmentBackInTheDay-20190329.pdf

The main changes this time are:

• page 11 - Added references to the use of HP-1000 machines for speech processing purposes
• page 12 - Inclusion of information about Mattel using Nuvatec tools including assemblers on the VAX
• page 12 - Addition of a section on Dr Color to the Mattel chapter
• pages 14-15 - Some details on the Magus architecture, implementation, memory size and tools. Plus the introduction of MEGAS as the more common name
• pages 15-19 - Updates to the Blue Whale test harness section, including feature evolution and corrections to the architecture
• page 22 - Section on Mattel's tools and stance on reverse engineering competitors' code

Last week I stumbled across this advert for the APh Datawidget from 1980, which I think is rather cool and I've added to page 7

 

(bonus marks for the first person that can tell us who currently is resident at 285 West Green Street and whether the phone is connected) Unfortunately, if anyone needs a summer job, although it looks as though you might have a few days to prepare, I think in reality you might have missed the opportunity

 

 

As always, all feedback is most welcome.

 

 

Cheers

decle

Edited March 29, 2019 by decle

[Utopia]

Looks like it is a carpark.

Edited March 29, 2019 by Utopia

[irvinetom]

Wow-amazing work, decle! And I must say, "a little technical consultancy" is putting it mildly—you and Joe are simply amazing. Braxton and I are at the National Videogame Museum again today, finding amazing stuff with the help of John Hardie. More updates to come!

[intvnut]

If I'm not mistaken, the little guy in the lower left corner is a Froy:

 

They make an appearance in a Frog Bog Easter egg. (Dabney eats it!)

 

 

As for APh's old digs, the closest I could get in streetview was 277 West Green Street. The building appears to be relatively new mixed-use. Probably whatever building was there is long since torn down.

 

[intvnut]

Also, I don't think I'll be giving these recipes a try. (From that same paper linked upthread.)

 

[+Lathe26]

On page 18, is says that Daniel's Black Whale ROM identifies as version 5f. Did we figure if my ROM card is the same? I think it was the same ROM but am not positive. There is a text message in my ROM that says "WELCOME TO 1610 DEVELOPMENT SYSTEM VERSION 5"

[intvnut]

On page 18, is says that Daniel's Black Whale ROM identifies as version 5f. Did we figure if my ROM card is the same? I think it was the same ROM but am not positive. There is a text message in my ROM that says "WELCOME TO 1610 DEVELOPMENT SYSTEM VERSION 5"

 

I'm pretty certain the one that decle and I analyzed is 5F. The "5" and the "F" come from different portions of the software, interestingly.

 

The CP-1610 side prints out the string "VERSION 5" (by sending it over a text FIFO to the KC EXEC), while the 6502 side prints out the "F" by poking a value directly into the frame buffer.

 

See the code at $8825 on the CP-1610 side:

.

        JSR     R5,     G_73D0      ; 8825   Print string
        DECLE   $0103               ; Set cursor column 3
        DECLE   $0143               ; Set cursor row 3
        STRING  "WELCOME TO 1610 DEVELOPMENT SYSTEM", $0D, $0A
        STRING  "   VERSION 5", 0

.

and the code at $E0B8 on the 6502 side:

.

E0B8  A9 46     LDA #$46    ; 'F'
E0BA  8D 4D B9  STA $B94D   ; Store to the frame buffer.

.

I'm not entirely sure why they did it that way; however my theory is that the separate reporting for 5 vs. F allowed each side to report its build revision (in case you somehow ended up with mixed EPROMs), and if one side or the other didn't show up, you'd know which side wasn't functioning correctly.

[+Lathe26]

 

I'm pretty certain the one that decle and I analyzed is 5F. The "5" and the "F" come from different portions of the software, interestingly.

 

The CP-1610 side prints out the string "VERSION 5" (by sending it over a text FIFO to the KC EXEC), while the 6502 side prints out the "F" by poking a value directly into the frame buffer.

 

See the code at $8825 on the CP-1610 side:

.

        JSR     R5,     G_73D0      ; 8825   Print string
        DECLE   $0103               ; Set cursor column 3
        DECLE   $0143               ; Set cursor row 3
        STRING  "WELCOME TO 1610 DEVELOPMENT SYSTEM", $0D, $0A
        STRING  "   VERSION 5", 0

.

and the code at $E0B8 on the 6502 side:

.

E0B8  A9 46     LDA #$46    ; 'F'
E0BA  8D 4D B9  STA $B94D   ; Store to the frame buffer.

.

I'm not entirely sure why they did it that way; however my theory is that the separate reporting for 5 vs. F allowed each side to report its build revision (in case you somehow ended up with mixed EPROMs), and if one side or the other didn't show up, you'd know which side wasn't functioning correctly.

 

That theory would be a clever way for them to detect mis-matched ROMs.

 

Thanks for looking at that code. Based off of this, I can confirm that mine is also a 5F, just from looking at the memory dump.

[ppelleti]

 

As for APh's old digs, the closest I could get in streetview was 277 West Green Street. The building appears to be relatively new mixed-use. Probably whatever building was there is long since torn down.

 

 

It looks like it was demolished between 1995 and 2002, as far as I can tell. (Interesting how the document says "assumed to be demolished", as if no one could be bothered to go check.)

[+nurmix]

I recently interviewed Mark Kennedy for a future episode of the podcast, and he talked about the Data Widget hardware debugger made by APh. It was the first I’d heard any of the BSRs talk about it.

 

 

 

 

Sent from my Keyboard Component using Jack's Conversational Intelli-talk cassette

[mr_me]

So, Mattel had a text to intellivoice speech conversion utility, with three different pitch options. Did anyone use this in a project.

 

I recently interviewed Mark Kennedy for a future episode of the podcast, and he talked about the Data Widget hardware debugger made by APh. It was the first Id heard any of the BSRs talk about it.

 

 

 

 

Sent from my Keyboard Component using Jack's Conversational Intelli-talk cassette

Whenever, you talk to soneone from the Atari Intellivision group see what they know about unreleased games e.g. joust, jungle hunt, pole position, missile command. Edited April 5, 2019 by mr_me

[+cmart604]

I recently interviewed Mark Kennedy for a future episode of the podcast, and he talked about the Data Widget hardware debugger made by APh. It was the first Id heard any of the BSRs talk about it.

 

 

 

 

Sent from my Keyboard Component using Jack's Conversational Intelli-talk cassette

Wait....there's going to be another episode? ??

[+nurmix]

So, Mattel had a text to intellivoice speech conversion utility, with three different pitch options. Did anyone use this in a project.

 

Whenever, you talk to soneone from the Atari Intellivision group see what they know about unreleased games e.g. joust, jungle hunt, pole position, missile command.

 

Hmm, not sure if you're joking with your first question (referencing my signature)... otherwise, no clue.

 

And yes, I did ask about that.

[+nurmix]

Wait....there's going to be another episode?

 

Eventually

[mr_me]

So, Mattel had a text to intellivoice speech conversion utility, with three different pitch options. Did anyone use this in a project.

 

What I meant is, did Mattel use the text to intellivoice speech conversion utility in any of their projects? I think all the intellivoice samples I've heard are from voice actors.

[intvnut]

What I meant is, did Mattel use the text to intellivoice speech conversion utility in any of their projects? I think all the intellivoice samples I've heard are from voice actors.

 

There's no evidence they productized anything based around text-to-speech in their products.

 

They were exploring multiple serial-terminal applications, and adding text-to-speech to terminal software was a popular past-time in the 80s. See TI-99/4A's Terminal Emulator II cartridge and the movie War Games. I think there may have been a text-to-speech terminal program that worked with the Apple ][ Mockingboard.

 

Also, text-to-speech seems interesting for BASIC programs (either on the Keyboard or ECS). I know TI BASIC supported it via the Terminal Emulator II cart. It's a way to allow the end user to have programmable voice above and beyond professionally recorded samples.

[+Lathe26]

... text-to-speech ... the movie War Games...

 

A bit off-topic, but I recently found out that the computer's voice ("Would you like to play a game") was not created via text-to-speech software. Instead, they had a human actor read the sentence backwards, they rearranged the words on tape back into the correct order (but now the normal human intonation is weird), and then ran it through some filters to give it a robot edgey sound.

[intvnut]

A bit off-topic, but I recently found out that the computer's voice ("Would you like to play a game") was not created via text-to-speech software. Instead, they had a human actor read the sentence backwards, they rearranged the words on tape back into the correct order (but now the normal human intonation is weird), and then ran it through some filters to give it a robot edgey sound.

 

That reminds me of the story of Stanley Kubrick and HAL-9000's voice. What that linked article doesn't cover (unless I missed it) was that Kubrick had acquired a nifty box that allowed him to speed up and slow down audio w/out changing its pitch. He used that to modify the cadence of HAL-9000's speech somewhat randomly, to give it an eerier, uncanny-valley feel.

[carlsson]

... and here we are today, trying to make our robotic voices sound as human like as possible.

[+nurmix]

What I meant is, did Mattel use the text to intellivoice speech conversion utility in any of their projects? I think all the intellivoice samples I've heard are from voice actors.

Gotcha. I really don’t know the answer, but I’ll ask the guys who would. And Joe... any idea if the built-in / canned phrases of the Intellivoice were all ‘voice acted’ or not?

 

 

 

...adding text-to-speech to terminal software was a popular past-time in the 80s. See TI-99/4A's Terminal Emulator II cartridge and the movie War Games.

I use the Odyssey 2 cartridge Type and Tell with The Voice unit for all the angry O2 bits in the podcast. And back in the day, I used SAM (Software Automatic Mouth) by Don’t Ask Software on my Atari 800XL. That had the ability to incorporate speech into BASIC code as well.

 

 

 

 

 

Sent from my Keyboard Component using Jack's Conversational Intelli-talk cassette

[mr_me]

The memo was dated 1983 june 21. I figure it was to be an inexpensive way for programmers to put speech in their games at a time when mattel was cutting down on costs. The intellivoice itself doesn't have the allophones so they would be loaded on the cartridge which is what the utility does, I think.

Edited April 6, 2019 by mr_me

[intvnut]

Gotcha. I really don’t know the answer, but I’ll ask the guys who would. And Joe... any idea if the built-in / canned phrases of the Intellivoice were all ‘voice acted’ or not?

 

 

I don't think we know who the voice actor is for the base Mattel voice, but I'm certain the RESROM phrases were provided by a voice actor and encoded and heavily edited.

 

The RESROM itself is highly structured to make all of the phrases fit. For example, "FOURTEEN" really is built out of two samples, "FOUR" and "-TEEN", while "FORTY" is built of "FOUR" and "-TY". All of the number samples were chopped up and remixed to match the reuse here. Frank Palazzolo (aka. Knarfian) did all the hard work of figuring out the phrase structuring within the RESROM. And he worked it out lightning fast, too. The dude's a genius, what can I say?

 

 

I use the Odyssey 2 cartridge Type and Tell with The Voice unit for all the angry O2 bits in the podcast.

 

The O2 allophone data is identical to the eventual SP0256-AL2 allophone data. So, if you use the AL2 allophone set we now include with IntyBASIC—with explicit permission from Microchip because I asked for it, I might add!—you're working with the same speech data.

 

For other samples on the O2, it's clear their voice encoding technology wasn't quite as sophisticated as Mattel's. Or, at least, they didn't spend as much time editing down and tweaking the samples. Mattel's samples are heavily edited and optimized, and it seems they had a particular emphasis on making the formants quite clear (almost to the point of being overemphasized). O2's samples have more of a "throw more ROM at it" feel, and aren't nearly as edgy sounding.

 

This difference in voice encoding philosophy also seems to fit with Mattel's complaints about how TI's samples sounded.

 

 

The memo was dated 1983 june 21. I figure it was to be an inexpensive way for programmers to put speech in their games at a time when mattel was cutting down on costs. The intellivoice itself doesn't have the allophones so they would be loaded on the cartridge which is what the utility does, I think.

 

They're not on the SP0256-012. They could be provided by a speech ROM, though, as the Intellivoice repurposes some cartridge port pins to bring the serial speech ROM bus out to the cartridge port. No cartridge (prototype or otherwise) has been discovered that utilizes these pins, however.

[mr_me]

I'm thinking the utility would make something that's useable on a cartridge, the same way the intellivoice cartridges have speech on them. Maybe it just used the GI allophones.

 

I think the voice actor who did the stuff inside the intellivoice also did some of the speech on the space spartans cartridge.

Edited April 6, 2019 by mr_me