Software Dectalk

[Tim Cross <tcross@xxxxxxxxxxx>]

I've spent the weekend playing with the version 5 of the software
dectalk from fonix. I've actually made some headway. I won't post the
sources just yet as they are'nt really fit for human consumption -
I've hacked them and filled them with debug comments and all sorts of
rubbish. However, I thought I'd communicate what I've found out and
possibly get some feedback/comments and maybe, for the really anxious,
they can just make the changes I've suggested to fix things up.


First off all, as reported on this list before, the version 5 of the
software doesn't seem to want to work with the existing dtk-soft and
tcldtk.c sources. A post from Gilles suggested some changes which
seemed to get things working. However, when I applied those changes, I
found the speech worked, but when the system stopped
speaking/synthesizing text, CPU load would increase to 100% and drop
to nearly nothing when it was synthesizing/speaking. This struck me as
odd and the reverse of what you would expect. 

It turns out, the problem is with the device option OWN_AUDIO_DEVICE
set in the tcldtk.c file. Once I removed that option and left the
devOptions value as 0, this problem goes away. So, the fix to get
tcldtk.c working with version 5 of the dectalk software is to use 

	unsigned int devNo = WAVE_MAPPER;
    DWORD devOptions = 0;
  
in the Tcldtk_Init() procedure. 

Once I got this working, the next problem was with the quality of
speech. From my playing around, it seems the problem is with the
in-line commands. I've made some minor changes with some of these and
have got some remarkable improvement. In particular, the section in
the dtk-soft script which deals with split caps (in the "clean"
procedure, seems to have been the main culprit. All of this has to be
taken with a grain of salt, as I am no expert on the in-line commands
and need to do some more reading/learning before I truely understand
the reason for some of them - Raman is definatly the expert with
these. However, I suspect, the software version of the dectalk is
unable to handle some of these in-line commands as reliably as the
harware dectalk. 

One symptom I was getting was speech drop-out and lots of variation in
volume. However, I modified the splitcaps handling section and now the
speech volume is better (and a lot better than the previous version of
the software BTW). Essentially what I have done is remove the in-line
dectalk commands in this section and replaced them with spaces. This
"seems" to work fine and the speech quality is a hell of a lot
better. 

So, if your using the software dectalk, you can either turn of split
capps support or make some modifications to the dtk-soft script. 

I will make my mods available when I get the chance to clean them up
and once I've checked to make sure I haven't introduced any other
nasties. I am now finding the Dectalk software synthesizer the best
all round solution I have found for a software synthesizer working
with emacspeak. 

Tim

-----------------------------------------------------------------------------
To unsubscribe from the emacspeak list or change your address on the
emacspeak list send mail to "emacspeak-request@xxxxxxxxxxx" with a
subject of "unsubscribe" or "help"