Dear Fellow Debian TTS gourp members, Debian a11y group members,

Recently Czech voices for festival have been offered for adoption.

As a founding member of the Debian TTS (text to speech) group which
maintains festival and speech tools
it suggests to me that the natural place to maintain Czech festival voices
is within the TTS group.

The advantages I can see is that:

1) Problems with voices are likely to be dependent on the changes in new
versions of festival as
voices evolve quite slowly in general.

2) TTS group have experience with the festival/speech-tools codebase and
thus have a bit
of experience with how festvox voices in the general sense are created and
how they function in the
Festival/Speech-tools framework.

The major disadvantage is that:

1) Currently the TTS team has no Czech speaker. Although the team is
fortunate to have speakers
of a number of different languages.

The question becomes how can Debian provide Czech voices for its community.
In the absence of a Czech speaker rapidly offering to take on the Czech
festival voices I make the
following proposal.

Although I am not a Czech speaker, I care enough particularly about fellow
Czech Debian community
to ask to bring Czech voices within the TTS group and work on it for the
Debian community.

I would be happy to integrate the voices into TTS groups alioth facilities,
monitors bugs lists, test
the voices with changes to festival/speech-tools.

I suspect that any problems with voices are likely to be all or nothing. It
works or its substantially
broken. Subtle problems might be detectable by people familiar with the
broad language family,
a Rosetta Stone (more on this later)  and sufficient motivation and a
desire to work with bug filers to
fix the problems.

The Czech lanaguage is partially intelligible to me (I am familiar with
another Eastern European language).

So what is this Rosetta stone I am suggesting?
In order for festival team members to look after the voices a set of
reference encodings would be generated
for the team which would be kept in their repository for testing purposes.
They would be moderate sized reference texts in the language and the
equivalent sound files rendered by festival voices.
This would guard against a fairly large range of faults in TTS generation
which would be obvious to
a motivated non-speaker. The more subtle faults could be checked with bug
filers and by a more
rare reaching out to the Czech Debian community for listening tests.

This effectively separates the need and interest in understanding how
voices are created and maintained
from the ability to native speak the language which would only require us
to find occasional listeners or have
them show interest through bug filling. My proposal intends to use the
limited resources of the Debian
community by separating interest in speech generation and technical aspects
with a outreaching
to native listeners who may not be interested in developing interest in
the technical aspects of voice maintainence.

I ask for comments from fellow TTS members, a11y team members and the
current maintainers for
on whether this is likely to be a workable scheme.
I also specifically ask if the retiring Czech voice maintainers would be
interested
in a short collaboration of creating
10-20 minutes of sound files from representative texts of Czech to act as a
Rosetta Stone for verifying
the voices during maintainence and to be included in our git repository.

Opinions Please?

best regards,
Peter Drysdale

Reply via email to