Newsgroups: comp.speech
Path: lyra.csx.cam.ac.uk!warwick!zaphod.crihan.fr!jussieu.fr!math.ohio-state.edu!cs.utexas.edu!csc.ti.com!tilde.csc.ti.com!cauldron!ra.csc.ti.com!usenet
From: nik@pluto.tiuk.ti.com
Subject: Re: SENSYN/KLSYN88 Group/Experts?
Message-ID: <CpyKAs.9zL@csc.ti.com>
Sender: usenet@csc.ti.com
Nntp-Posting-Host: pluto.tiuk.ti.com
Organization: Texas Instruments Ltd.
References:  <1994May17.124327.23793@ucl.ac.uk>
Date: Tue, 17 May 1994 17:52:04 GMT
Lines: 53

     Mr Andrew Simpson wrote in article <1994May17.124327.23793@ucl.ac.uk> :
>
>
>
>Is there a user-group for the KLSYN88/SENSYN Klatt Synthesiser?
>
>If not, is there enough interest to warrant starting one?
>

I am not sure how KLSYN88/SENSYN relate to Jon Iles/my version
but I am certainly interested in keeping in touch with anyone
using Klatt-derived synths.

>We're using the Klatt to produce copy-syntheses of natural speech
>tokens, and would like to know if there are any pitfalls, known
>shortcomings/bugs in the software, and if there are nuggets of arcane
>information known only to a few experienced users out there.
>
Some of it is certainly arcane! I am currently trying to upgrade
"rsynth" to use cascade/parallel mode, and I am having problems.
There are various "constraints" which are automatically applied 
by (for example) earlier MITalk modules which are vital for sensible
output. The appendix of the MITalk book which describes these could
win an award for non-lucidity ;-).

>For instance, what is the optimal number of formants for various
>sampling rates?  Presumably as the sampling rate increases, so should
>the number of formants. 
>
Number of formants is as I understand it characteristic of the "speaker".
So once sampling rate is "sufficent" to cover all the formants
needed for a "speaker" you don't add any more.

Certainly you should NOT include formants with centre
frequencies above Nyquist rate for sampling rate. Whether you should
adjust bandwidth so "skirts" don't exceed Nyquist rate is moot.
I have certainly "heard" problems when enabling formants where 
f+bw/2 > Nquist, but have not figured out causes.

>Additionally, does anyone know anything about a possible bug in the
>radiation characteristic applied to voiceless excitation on Unix/VMS
>systems?
>
Not knowing which "source" you have I can't say for sure, but I would
say it is quite likely. 

Klatt says in (some versions of) descriptions
that radiation characterisic can be folded into voicing source.
If this has been done and actual radiation characteristic 
filter is removed you may well get a mess which is only manifest
for voiceless excitation (unless that is hacked to compensate).

Nick Ing-Simmons
