Newsgroups: comp.speech
Path: pavo.csi.cam.ac.uk!warwick!uknet!bnr.co.uk!pipex!uunet!cs.utexas.edu!csc.ti.com!tilde.csc.ti.com!fstop.csc.ti.com!pluto!a904209
From: a904209@server1.tiuk.ti.com (Nick Ing-Simmons)
Subject: Public Domain synthesis
Message-ID: <C8yt9J.Jtq@csc.ti.com>
Keywords: klatt, British
Sender: usenet@csc.ti.com
Nntp-Posting-Host: pluto.tiuk.ti.com
Reply-To: a904209@server1.tiuk.ti.com
Organization: None
Date: Mon, 21 Jun 1993 09:34:30 GMT
Lines: 34

I am new to to speech synthesis, but have a lot of experience as
a C programmer.  I have pulled the "klatt" package from the 
comp.speech archive site and have had some success at getting it 
to work on my Sun IPC.

In particular the "backend" (sklatt) package produces a reasonable
version of the "hello.par" file that was part of the package.

However, the "gen" (phoneme to parameter) program seems to be in
poor shape. As a C programmer there are a few "obvious" bugs of like 

if (!features & VOWEL)  
rather than 
if (!(features & VOWEL))
or perhaps 
if (~features & VOWEL)                                     

!features means (features != 0)

Would any of the authors like to correspond on cleaning this up,
with a view to re-posting a "professionally" coded version?

Is there a book or paper which describes the "gen" algorithm?

Finally does anyone know how the synthesis processes
should be changed to get "British" rather than "American"
accent ?  Are the "phonemes" the same, but differently parameterized,
or are there really other vowels.

Please note that although I am using my employer's Internet connection
to post this, this is a personal project.

Nick Ing-Simmons   nicki@lobby.ti.com

