Build a speech generating device

[Hugh Sasse]

On Sat, 30 Jul 2011, Frederik Elwert wrote:

> there are only few speech generating devices (SGDs) available on the
> market, and those are as limited as they are expensive, I plan to build
> a custom SGD using a tablet computer as a basis and applying available

I don't know about the hardware.  I knew people who used Texas
Instruments chips, but I don't expect they'd include all the German
phonemes.

> The primary components I identified to be necessary are
> 
>       * a virtual keyboard with word prediction 

OK, the only one I know of is Dasher, which you have found.  
The inference group have a thing called Tapir, which is designed for
on screen text entry like "texting".  I don't think it will do 
all the symbols on the keyboard, but it is at
http://www.inference.phy.cam.ac.uk/tapir/
It didn't seem to be a quick way of getting text in, but for people
with low mobility it may have some use.

In the book "Beautiful Code" [Andy Oram, Greg Wison, O'Reilly, 2007,
ISBN:9780596510046]  Chapter 30 "When a button is all that connects
you to the world" discusses the software used by Professor Hawking.
It claims the download is at
http://holisticit.com/eLocutor/elocutorv3.htm 
although I can find nothing useful there.  The search engines take
me to
http://hawking.sourceforge.net/
and it appears that the download is available as an executable or a
Zip file, so I suspect it is Windows only.

For prediction there is also Presage 
http://presage.sourceforge.net/
which is really a library, so could be attached to something else.
It does have some wxPython demos, which I can't get working [on
Cygwin], though your experience on Linux could well be better.

Way back, there used to be a program called reactivekbd, which was
a predictive text entry system that could be used from the shell.

It seems to be here:
http://ftp.sunet.se/pub/usenet/ftp.uu.net/comp.sources.unix/volume20/reactivekbd/
I had that working under Sunos 4.<mumble>, but have not retried recently.

The dasher project does have the Tcl/Tk dasher which may still be
useful if you can't get the rest to build and work, but that ought to
reasonably easy to connect to the HTTP interface of OpenMary.

>       * pre-defined text snippets 
>       * a speech synthesizer backend (for German language output) 

I think espeak supports German, but I'm not in a position to
comment on the quality.
http://espeak.sourceforge.net/

>       * a frontend to the speech synthesizer

Both OpenMary and Espeak have front ends you can type into.  The
OpenMary example client is in Java, and there is a Ruby one and a Python
one in the repository now.  They will need more work for non-Windows:
lots of choices for sound on Linux.

> 
> For the speech synthesizer, I currently plan to use OpenMary[1], since
> its output quality is significantly better than espeak?s, even with
> mbrola voices.

I think they are dropping mbrola voices because they need a non-java
backend for it, and they mostly have prosody working now.
See Msg Id: 4DA410AD.8040403 at dfki.de posted to Mary-users on 12 APR 2011.

> 
> For the speech synthesizer frontend, I plan to either adapt gespeaker,

I don't know about gespeaker, so I searched, and found this:
http://alternativeto.net/software/gespeaker/
thereby finding Kmouth
http://www.schmi-dt.de/kmouth/index.en.html
which claims to have word completion and a phrase book, as well as
history.

        [The rest trimmed]

        Hope some of that helps,
        Hugh