[DECtalk] DECtalk on old phone system

[Don]

Hi Blake,

On 2/16/2021 1:45 PM, Blake Roberts wrote:
> I am in fact interested in how Dectalk was used in addition to firmware sound.

I will put "look for old firmware" on my list of things to do.  I am presently
busy rebuilding several of my workstations so the office is resembles the
aftermath of a tornado...

Speech is just one modality that can be used to communicate with users.  Its
most obvious application is for the visually impaired (assuming not also deaf).
Or, for applications where the medium doesn't support the bandwidth needed
for video.

I'm more interested in schemes for designing applications that are "output
neutral" -- so, the same application can adapt to the modalities most
applicable to the current user(s).

For example, an alarm clock application shouldn't have to be rewritten or
redesigned to deal with a sighted/blind, hearing/deaf, ambulatory/paralyzed
user.  The nature of the problem being solved (for example, notifying the
user when a particular time has come) shouldn't have to be changed to
accommodate the user's sensory/interactive needs.  Otherwise, you end up
with applications that target just particular subsets of the potential
user base instead of being universally accessible.

> I find it fascinating how Dectalk can be useful in so many situations. Example:
> I remember that years ago a member of this list helped make Dectalk speak a
> choice-based game on the TellMe phone system. This was back in the day when the
> service offered extensions. I would call TellMe and enter the extension number
> just so I could hear/use Dectalk.

IVR is considerably more commonplace, nowadays.  Everyone has a bank account
that can be accessed over-the-phone delivering up-to-the minute balance
information.  Doctors' offices deliver personalized reminders of upcoming
appointments annotated with your name, date and time.  Public library's remind
you of overdue books.  etc.

Most of these -- and, I suspect, the game you described, above -- are limited
domain applications; you don't need to be able to utter ANY possible sequence
of words (comprised of arbitrary sequences of letters, numbers and symbols).
As such, you can get much more natural speech addressing just those necessary
utterances instead of trying to accommodate any arbitrary word sequence (which
may or may not even be gramatically correct).

Aardvark beans green running zero diligently?

Here, for example, I can inform you that the garage door is "open", "closed",
"opening", "closing", "inoperative", "jammed", "has detected an obstruction
in its path", "almost open", almost closed"... but, not that it is "white" or
"warm to the touch".  You may be interested in those facts.  But, I can't
sense them and, thus, why bother trying to report on them?

Likewise, I can tell you that the weather outside is "warm", "sunny", "cold",
"windy", "raining", "snowing", "a mixture of sleet and rain"... but, not
"delightful for a picnic" or "reminiscent of an early spring evening".

Similarly, I can indicate the nature of system failures without resorting
to having to utter human-unfriendly sequences of technospeak:  "Mutex
deadlock in thread chk_door_stat() at 0x0344FED9" -- which would be
silly to expose to a user, regardless!

So, the value of unconstrained synthesis is greatly offset by the desire for
intelligibility.

Thinking about what you want to convey also addresses how you will likely have
to convey that in a different modality:  what if the user is deaf?

> Also, for a brief period Dectalk was used on the National Federation Of the
> Blind's NewsLine service if I recall correctly. This was back before eloquence
> was utilized instead.

IIRC, Kurzweil's Personal Reader (and, perhaps, the Model 3) also used DECtalk
instead of the Votrax units used in the early models.  But, there you are
faced with truly unconstrained input as the text you're parsing can be
damn near anything!