[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Speech Generation/Recognition software for 3600s

I am looking for any available Speech Recognition or Speech
Generation software which runs on Symbolics 3600s. I need
both generation as well as recognition so any pointers to
either would be helpfull. As far as generation goes something
similar to Macintalk (on the Macintosh) is ideal. It should
use the builtin sound generation capbilities available on the
3670/3640. It will not be used for English so English text to
speech pronounciation rules etc. are not needed. Only the ability
to convert some internal phoneme representation to audible sound
is needed. Ideally there should be some way of controlling the
parameters of the phoneme set though anything capable of producing
the English phoneme set will pretty much do. As far as recognition
goes, again I it will not be used for English so I only need
conversion from sound to some internal phoneme representation.
Ideally it should allow continuous speech though I could make
do with discrete speech if necessary. It should have as large
a vocabulary as possible (ideally infinite if it just gives me
phonemes). Those are the critical parameters, the noncritical
ones are as follows. I don't care about how long it takes to
train, it could take forever (almost). I don't care if it is
speaker dependent. I don't care if it is not real time, up to
three or four times slower than real time. As there is no builtin
A/D converter in the 3600 I will need some additional hardware
but I would like to keep this to a minimum. I have looked into
Kurweil's system as well as Speech Systems Technology's "Phonetic
Engine" (TM). Neither hooks directly to a 3600 and must go through
an IBM PC to the serial port on the 3600 console. Kurzweils system
is discrete speech and has only a 1000 word vocabulary. More
importantly, it fails on monosylabic words which I need to recognize
accurately. The phonetic engine seems ideal but is not realy "there"
yet. The demo I saw did recognize continuous speech, but only with
a 40% hit rate. Anyway, I would prefer some minimal hardware solution
if possible.

I would appreciate any pointers to any software, (preferrably public
domain or otherwise free), but am nontheless interested in ANY pointers
including such things as source code in other languages (like for Macintalk),
algorithm descriptions, commercial vendors, etc.

Send replies to:
Jeffrey Mark Siskind