[orca-list] Speech system shootout

["Nolan Darilek" <nolan thewordnerd info>]

Hi, all.

I've been  using the Orca Speech-dispatcher support for a while, and 
recently decided to play with the other choices, wondering how 
performance and such differed. I've had mixed experiences with each and 
thought I'd share them. Since each behaves well or poorly in a variety 
of ways, I'd like to file bugs but don't know where to file them.

Gnome-speech with Speech-dispatcher: This seems a bit more responsive 
than my initial SD solution, and I find that I can type ahead of 
characters as they're speaking, which is refreshing. The first thing I 
encounter not working, though, is say all. It seems to read one element 
in Firefox--paragraph, header, whatever, then stop. This isn't true in 
the Thunderbird message entry window or in Gedit. I also couldn't get 
the voice to change at all--it remains a British male, even when set female.

Gnome-speech with Espeak driver: This has the best say all performance, 
reading everything very quickly. It also stops the cursor very close to 
where I stop reading. I have to cut my test short, though, because it 
doesn't interrupt itself when I type, making my first attempt at typing 
this paragraph painful.. I also seem to recall it getting interrupted 
during a say all by an IM and just stopping, which is rather frustrating.

Native speech-dispatcher: This behaves the closest to what I'd expect a 
screen reader to do, though it has its quirks. Say all doesn't stop as 
happens with Gnome-speech+SD, nor is it interrupted by new text as is 
true with Gnome-speech+Espeak. Rather, new text gets inserted neatly 
into the flow, albeit sometimes not as immediately as I'd like. It's 
also very responsive for typingg. It has its quirks, though. I've 
noticed a commonality in both SD drivers in that there is a noticable 
pause for punctuation, even when none is spoken. For instance, run 
spd-say with the following test strings:

"This is a test."
This is a ( test ) ."

This part is obviously an SD issue, and I'm not immediately sure how to 
resolve it, other than to check whether all the contents of a string are 
beneath the punctuation threshold of the currently selected mode and are 
non-alphanumeric, and if so remove them from the string. I notice this 
behavior at other times, too, like before links are spoken. There is 
often a pause of nearly a second before the word "link" speaks. There 
are also other inconsistencies. " is "double quote" when typed, but 
"quotes" when backspaced over. Some of the character names are just 
plain strange. ' is "quote", " is "quotes," but only when erased, see 
above. :) ( is "left bracket," [ is "left square," { is "left brace." 
Even after nine months of use, I still have to pause and ask myself 
which is which as this is inconsistent with every other speech system on 
every other OS I've used, and that isn't because each is inconsistent 
with the other. :) The SD driver doesn't seem to stop as smoothly, 
either, and usually I have to find my place when concluding a read. 
Despite these quirks, though, I'd say that for an experimental driver, 
it is closest to what I'd expect from  a screen reader.

So why dump all this on the Orca list? First, because it is the 
user-visible entry point into each of these systems. Next, because I 
don't know where some of these issues should be solved and am hoping 
that someone here has a clue. :) If both Gnome-speech drivers didn't 
handle say all correctly, I'd conclude that Gnome-speech or Orca might 
be at fault, but one of them handles it fine while the other doesn't. 
Same deal for interruption of typing, both drivers in the same system 
are inconsistent. I suppose it could be issues with the drivers 
themselves. In that case, I'd be interested in reading others' 
experiences with these systems, workarounds, etc. And,  again, I'm happy 
to file bugs if we know to where they should be sent. :)