[DECtalk] DECTalk 4.2 Alpha (fwd)

[Don]

On 8/16/2022 5:19 AM, joshknnd1982 at gmail.com wrote:
> The votrax SE02, the one used in many notetakers and speech synthesizers. Is
> there any emulators for that one?

I assume you mean the SC-02 aka SSI-263.

Votrax's approach to synthesis was entirely analog.

The original boardset consisted of 4 (?) sizeable "modules"
(about 2.5 x 6" each) *potted* to hide their contents (of
course, that's not an effective way of protecting a design as
anyone can opt to remove the encapsulation and then it's
just a set of circuit boards using COTS components!)

The modules were color-coded: red, green, beige and grey.
They served different functions: delay, fricative, vocal
tract and transition.  Optional modules provided a serial
interface and "stored phrases" (via ROMs).  You could also
hook up a (special) keyboard to control the unit without
a computer being present.

The whole unit was a bit larger than a USFF but a bit smaller
than a SFF PC.  Cost was ~$1000.

All of the waveform generation was done with analog circuitry.

Also note that the device only created "phonemes".  It was
the responsibility of the device driving it (e.g., a computer)
to figure out which phoneme to utter and in what sequence,
how to apply prosody, etc.

Devices like the Type n Talk, Personal Speech, Microvox,
Intex Talker, etc. augmented the "waveform generator"
(Votrax) with a small microprocessor that applied a canned
set of rules to determine which phonemes/intonation to
generate to render a particular set of letters into speech.

The algorithm used seemed to be the NRL ruleset with, possibly,
some minor optimizations (exception word dictionaries, etc.)
AFAICT, the only competing "simple" letter-to-phoneme ruleset
available at that time was McIlroy's.

[Most of these were remarkably bad -- estimates of ~60% accuracy
were not uncommon -- as they couldn't do much high-level parsing
of the input, complex sentence analysis, etc.]

The VS6 could sing (if you manipulated the data sent to it,
appropriately -- on a per-phoneme basis!).  I heard a VS6.3
speak convincing German!

With that as background, the SC-01 and SC-02 were "integrated"
versions of the same basic technology.  Switched capacitor filters
were used to get everything onto a chip.  But, the voice quality
was largely the same.

We used to joke that you needed 12 inches of concrete to block
out the sound of the voice!

By contrast, DECtalk (KlattTalk, MITalk, etc.) is a pure software
approach.  The filters/resonators are implemented in the digital
domain with the resulting "speech" happening at the very end of
the process where the DIGITAL waveform is converted to analog form.

Also, the letter-to-sound rules are implemented *in* the synthesizer
instead of being the responsibility of some external "program".

While it could be possible to write an emulator for one of the
votrax designs (i.e., map each analog filter into a corresponding
digital/software implementation and the various digital controls
of those filters into their corresponding equivalents), I can't
see any value to doing so other than as a monument to oneself
("I didn't have anything important to do so I've resurrected
an obsolete technology instead of buying an SSI-263 on eBay")