[hts-users:01490] Re: hts_engine-0.99 does not generate normal speech in HTS-demo using 12-th order LSP

[Sacha Krstulovic <sacha@xxxxxxxxxxxxxxxxx>]

    Hello Heiga;

Thanks for this clarification.

> Yes, we changed the header of GV pdf in hts_engine API 0.96. [...]
> Similar discussions have been made in hts-users:01300, 1302, 1303, and 
> 1304.

Sorry I didn't make the connection with the endianness topic. :o)


> GV pdfs are generated by HTS-demo/data/Makefile, not HTS itself.
> So I think hts_engine API and HTS are independent.
>
> Sometimes we have to change model format etc.
> I recommend you to re-train your model using the latest HTS
> and demo script because the latest one may have less bugs.
>
> # I personally believe that the latest HTS is always the best one :-)

That's for sure, no question about the latest version being the best.

However: I have seen many messages in this list where people got
confused with old models or old HTS demos mixed up with new versions
of HTS itself (or in my case a new version of hts_engine).
Do you think something would need to be improved there? E.g:

From your answer, it is clear that a particular version of
HTS-demo makes the link between a particular version of HTS
and a particular version of hts_engine. The configure script
of the HTS-demo does check the version of hts_engine, but does
not seem to check the version of the HTS modules, so a confusion
is still possible. Maybe the -V switches for various HTS modules
could return some HTS version info in addition to HTK version
info (they don't seem to do it for HTS-2.1beta), and the configure
of HTS-demo could check it the same way it checks for hts_engine?

Also, even with the demos checking for HTS and hts_engine compatibility,
it is still possible to mismatch some readily trained models with
a wrong version of HTS or hts_engine (the way I did).
Would it be a good idea to add a version ID in the format of
the models, to check for compatibility or detect some version
mismatch problems faster?
Or, alternatively, should hts_engine_API come back to following the same
version numbering scheme as HTS+HTS-demos to avoid such confusions?
Or should compatibility information be clearly referenced somewhere,
e.g., on the website or in the README files of the various
modules?

I understand that for the moment, all the latest versions
of HTS, HTS-demo and hts_engine_API should be used together,
but if the plan is to allow these packages to be developed
and distributed independently, then at some point the projects
will face such compatibility problems.

You may also want to consider that it can take some time for
a lab to migrate to a new version of HTS if they have had
to tailor some training scripts to their own data, their own
computing resources and/or their own language.  So it is difficult
to expect everyone to run the latest version in a short time span,
besides running the delivered demos out of the box; from that
point of view and as the use of HTS becomes more and more widespread,
it may be useful to help the users identify such compatibility
issues more clearly.

I guess many of us are just waiting for the best version,
HTS-2.1 and hts_engine_API-1.0! ;-)

I hope some of these suggestions help, and thanks a lot
to the HTS team for their development work;
                                            -*- Sacha K. -*-
--
Dr.Sacha Krstulovic - Research Engineer
Toshiba Research Europe Limited
Cambridge Research Laboratory
Speech Technology Group
208 Science Park, Milton Road
Cambridge CB4 0GZ - United Kingdom
Tel:    +44 1223 436 978
Fax:    +44 1223 436 909
E-mail: sacha@xxxxxxxxxxxxxxxxx



______________________________________________________________________
This email has been scanned by the MessageLabs Email Security System.
For more information please visit http://www.messagelabs.com/email 
______________________________________________________________________
begin:vcard
fn:Sacha Krstulovic
n:Krstulovic;Sacha
org:Toshiba Research Europe Limited - Cambridge Research Laboratory;Speech Technology Group
adr:;;260 Science Park, Milton Road;Cambridge;;CB4 0WE;United Kingdom
email;internet:sacha@xxxxxxxxxxxxxxxxx
title:Research Engineer
tel;work:+44 1223 436 978
tel;fax:+44 1223 436 909
x-mozilla-html:FALSE
url:http://www.toshiba-europe.com/research/crl/
version:2.1
end:vcard