[mary-users] German voice proprietary? Clarification needed.

[Klaus Knopper]

Dear Marc,

On Fri, Oct 05, 2007 at 12:50:00PM +0200, Marc Schroeder wrote:
> Dear Klaus Knopper,
> 
> first of all, thank you for creating Knoppix -- an excellent tool that I 
> have used many times.

Thanks for testing. Knoppix, which like many other open source projects,
naturally lives from feedback and code contributions from the community.
:-)

> As the main developer and team leader of MARY TTS development at DFKI, I 
> will try to answer your questions.

Thanks for taking the time. I apologize in case my previous mail
sounded very strict about open source as a mandatory requirement for TTS
in my opinion. Well, but in fact, it is my firm conviction that a TTS
for public use, for a variety of reasons, MUST be open source, in its
entity including voices.

Of course, it would be extremely cool to have freely distributable voice
databases for different models (diphone, half-syllable, unit
selection...) that are somewhat interchangeable between different OSS
TTS engines of the same model.

> In essence, your perception of the current license issue is correct -- 
> there are two parts:
> 
> * OpenMary is open source, under a BSD-style license; it encompasses the 
> core of the system, including the unit selection code, the voice 
> building and signal processing tools, as well as the Tibetan and most of 
> the English modules;

Same as festival, the engine itself is open source and not restricted in
distribution or fields of use, but there are not many free voices
available. Unfortunately, also for festival there is no freely
distributable german voice available (mbrola, which is supported in the
way of a kind of plug-in, suffers from the same problem of not being
distributable in an unrestricted / commercially applicable way).

> * DFKIMary is the part that we cannot make open source for a variety of 
> reasons; the best we could do for the community was to release it for 
> download under a research license. It encompasses the German modules and 
> voices, and the English part-of-speech tagger.
> 
> In other words, you are free to use OpenMary in whichever way you like; 
> to re-distribute, change, etc., even sell. But DFKIMary you cannot reuse 
> in this way. So indeed, on the basis of the current license, you cannot 
> distribute a German-speaking MARY TTS system, but an English-speaking 
> system is no problem. This qualifies your interpretation that "the 
> software as a whole is non-distributable".

This is what I assumed. It's sad, so much work done on languages and
exensions that will never make it into the public, and that will never
be used in practical everyday appliances. Few people will ever know that
there was a functional german voice of good quality, just as a research
project.

> The choice of a BSD-style license for OpenMary was motivated by the wish 
> to be compatible with both "free" (in the sense of the GPL) and 
> closed-source (as in DFKIMary) code. Within the limits of our 
> constraints, this combination was the most liberal we could do.
> 
> What kind of time and work would you have liked to invest into the 
> development and evaluation of the MARY TTS system?

In summary, we are building a menu-driven system that is usable by blind
computer-beginners (not experts of the commandline), and also for people
who are for other reasons unable to read what's on the screen, so we
need a multilingual, completely distributable and also
commercial-appliable open source speech engine with distributable voices
in as many languages as possible. Since computer-beginners are involved,
the voices should also sound "natural", even that for many people who
are USED to speech synthesizers, understandability of fully synthetic
voices would be sufficuent.

Creating our own voice recordings for festival (duplicating the efforts
that have already been done for mbrola and others), would slow down
development too much and increase costs, cosidering that we also would
like to support other european languages. Our best choice so far is
espeak, which is purely synthetical, but with its small size of about
60k (including 30 different voices/languages) even fits into a bootstrap
loader. But we will continue evaluating other OSS TTS systems as well.

> If you have a concrete use in mind, I suggest you contact me directly 
> (not via the list), and we can discuss possibilities for providing 
> DFKIMary components under a different license for a specific purpose. I 
> am not promising anything, but discussing options may be worth the time.

I would be happy to do this. For an overview of what we are doing,
please have a short look at
http://knopper.net/knoppix-adriane/index-en.html .

With kind regards
-Klaus Knopper