Re: SP0250 data format

At 10:41 AM 6/12/97 -0800, you wrote:
>>At 09:50 AM 6/12/97 -0800, you wrote:
>>The SP0256 was GI's formant synth version of the chip. They simply took the
>>SP0250 added an internal ROM of LPC recorded formant phrases for 56(?)
>>different parts of speech. Then instead of accessing full phrases, you
>>accessed formants and strung them together to create phrases.
>
>Seems like we've gone over this before... Anyway, this is the message from
>Larry to me. (My quotes are double ">>", Larry's are ">")...
>
>>>I have a kind-of trivia question for you. Since you've been doing voice
>>>compression for quite a while I wonder if you ever ran across or knew
>>>anything about the General Instruments SPO-250 "Orator" chip. The
>>>SPO-256AL2 was their little phoneme synth that was used in quite a few
>>>gizmo's of the early 80's (maybe the Intellivision Voice module I think),
>>>but the Orator was used in some Sega Arcade games (like Star Trek) and
>>>sounded *very* nice for the era. I assumed that it was some form of LPC
>>>since the voices didn't really sound digitized like an ADPCM or something.
>>>Very faint "robotic" overtones sometimes on vowels... Any thoughts?
>>>
>> I definitely remember the name and number, and I'm pretty sure that I've
>>seen a spec sheet for it. The memories are faint (and not just about speech
>>stuff!), but I think this was a formant synthesizer. The SP0256 was
>>essentially the same part preprogrammed with a phoneme/allophone set. At
>>the time, formant synthesizers were the clear favorites for doing
>>text-to-speech. Conventional wisdom was that "they" would have the kinks
>>worked out of TTS in short order and then there would be no need for voice
>>coders- at least for playback-only systems. Co-incidentally, that was about
>>the time one of my partners here started on his 8-year TTS research project
>>at CNET in France. In spite of being one of the world's best systems at the
>>time, neither it nor the others have ever really been good enough for the
>>big time-- Michel refuses to touch the stuff now!
>> Back to the 0250; the problem for formant synthesizers in that period was
>>(the lack of) automatic formant tracking. As it happens, the most popular
>>filter control parameters these days are line spectrum pairs (of
>>frequencies- LSPs) which come pretty close to tracking what a shadetree
>>like myself would consider to be formants, though this time around, no one
>>pretends that there's a one-to-one relationship. Sorry-- got carried away
>>there.....

Whoaaaa! Lost me there also!

I just talked to a guy here at work that knew a little about these things.
The way he described it is: Formant synthesis uses pitch control and filters
to try and re-create the vocal cords and vocal track, whereas LPC was more
of a compression technique that is designed around the many rundancies found
in speech.

Given that it appears the Orator was definitely a "formant synthesizer".

I on the otherhand was confusing "formant" with "phoneme" which according to
Larry's letter (and my co-worker) are quite different.

If anything this seems to make recording new sounds for the Orator even a
more difficult task, since I believe finding algorithms for LPC would be
easier than knowing the makeup of the filters, and oscillators used in the
GI chip, and the mappings used to access them.

-Zonn
Received on Thu Jun 12 11:12:13 1997