In this chapter we will try to list some of the important resources
available that you may need when building a voice in Festival. This
list cannot be complete and comprehensive but we will to give references
to meta-resources as well as direct references to information code, data
that may be of use to you.
This document itself will be updated occasionally and it is worth
checking to ensure that you have the latest copy.
Updates, new databases, new language support etc will happen
intermittently, new voices will be released which may help you
develop your own new voices.
http://festvox.org
has been set up as a resource center for voices in Festival offering
databases, examples and repository for voice distribution. Checking
that site regularly is a good thing to do.
Specifically
- http://festvox.org/examples/cmu_us_kal_diphone/
Offers a complete example US English diphone databes as built using the
walkthough in Chapter 19. The originally
recorded diphone databases is also available as is, at
http://festvox.org/databases/cmu_us_kal_diphone/.
- http://festvox.org/examples/cmu_time_awb_ldom/
Offers a complete example limited domain synthesis database as
build using the walkthroughs in Chapter 5.
Other databases, lexicons etc will be installed on festvox.org as
they become available.
There is also a mailing-list [email protected] for
discussing aspects of building voices. See
http://festvox.org/maillist.html for details of joining it
and the archive of messages already sent. Also, while traffic is low,
feel free to mail the authors [email protected] or
[email protected] and we will try to help where we can.
The Festival home page http://www.cstr.ed.ac.uk/projects/festival/ It is updated
regularly as new developments happen.
The Festival Speech Synthesis System code and the Edinburgh Speech
Tools library and related programs are available from
ftp://ftp.cstr.ed.ac.uk/pub/festival/
or in the US at
http://www.festvox.org/festival/downloads.html
Note that precompiled versions of the system are also available
from that site, though at time of writing only Linux binaries are
available.
Festival comes with its own manual and html, postscript and
GNU info format. It and a less comprehensive Speech Tools
manual are pre-built in festdoc-1.4.1.tar.gz. The
manuals are also available on line at
http://www.festvox.org/docs/manual-1.4.2/festival_toc.html
http://www.festvox.org/docs/speech_tools-1.2.0/book1.htm
You will likely need to reference these manuals often.
It will also be useful to have access to other voices development
in Festival as seeing how others solve problems may make
things clearer.
In addition to Festival itself a number of other projects throughout
the world use Festival and have also released resources. The
"Related Projects" links give urls to other organizations
which you may find useful.
It is worth mentioning Oregon Graduate Institute here who have done a
lot of work with the system and release other voices for it (US English
and Mexican Spanish). See http://cslu.cse.ogi.edu/tts/ for more
details.
A second project worth mention, is the MBROLA project [dutoit96]
http://tcts.fpms.ac.be/synthesis/mbrola.html, they offer a
waveform synthesis technique [dutoit93] and a number of diphone
database for lots of different languages. MBROLA itself doesn't offer a
front end, just phone, duration and F0 target to waveform synthesis.
(However the do offer a full French TTS system too.) Their diphone
databases complement Festival well and a number of projects use MBROLA
databases for their waveform synthesis and Festival as the front end.
If you lack resources to record and build diphone databases this is a
good place to check for existing diphone databases for languages. Most
of their databases have some use/distribution restrictions but they
usually allow any non-commercial use.
