[Kde-accessibility] ktts - literature

[Gary Cramblitt]

On Wednesday 22 December 2004 03:24 pm, Mikolaj Machowski wrote:
> 1.
> I am using Festival plugin and tried to put one of books
> from Baen through ktts.
>
> I don't know where exactly lies problem - KTTS or Festival but
> there is problem with properly cutting in sentences.
>
> Try this fragment:
> (first sentences of 'Diplomatic Immunity' by Lois McMaster Bujold)
>
> "Once more, dear boy, into the breach - for England, Harry, and Saint
> George!" Miles murmured encouragingly. "Or at least, for Barrayar, me,
> and maybe Grandfather Piotr. Ha!" With a last twitch, the sperm vanished
> within its destined paradise.  "Miles, are you looking at those baby
> pictures again?" came Ekaterin's voice, amused, as she emerged from
> their cabin's sybaritic bathroom. She finished winding up her dark hair
> on the back of her head, secured it, and leaned over his shoulder as he
> sat in the station chair. "Is that Aral Alexander, or Helen Natalia?"
> "Well, Aral Alexander in the making."
> "Ah, admiring your sperm again. I see."
>
> Note lack of pauses in:
> Ha!" With
> again?" came
> Natalia?" "Well
> making." "Ah
>
> There is problem with proper treating of " and interpunction signs.
> This can be quickly solved by removing " when piping text into Festival.
> This is rather hack than real solution but it works, try this:
>
> Once more, dear boy, into the breach - for England, Harry, and Saint
> George!  Miles murmured encouragingly.  Or at least, for Barrayar, me,
> and maybe Grandfather Piotr. Ha!  With a last twitch, the sperm
> vanished within its destined paradise.  Miles, are you looking at those
> baby pictures again?  came Ekaterin's voice, amused, as she emerged
> from their cabin's sybaritic bathroom. She finished winding up her dark
> hair on the back of her head, secured it, and leaned over his shoulder
> as he sat in the station chair.  Is that Aral Alexander, or Helen
> Natalia?
> Well, Aral Alexander in the making.
> Ah, admiring your sperm again. I see.

I want to investigate this some more.  Interestingly, your second example is 
actually worse when fed directly into Festival in a konsole via command 
"festival --tts sample.txt".  What you are talking about is Sentence Boundary 
Detection (SBD), and its a difficult problem for synths.  Both KTTS and 
Festival perform SBD.  KTTS does it so that you can rewind, advance, skip 
sentences, etc.  As you've observed, KTTS doesn't properly handle inline 
quotations, which I'll try to address in the future.  Festival also does SBD 
so it can insert pauses and has some anamolies of its own.  The reason why 
your second sample sounds better when fed to KTTS is KTTS does better SBD 
with the quotes removed.

> 2.
> How to add new language support to KTTS?
> There is a bit tricky way (manipulation of Welsh) to interpret Polish
> texts in Festival and I'd like to try to add it.

Typically, TTS takes place in two phases.  1.  A text to  phoneme phase, and 
2.  Phoneme synthesis phase.  I'm guessing you have a txt2pho program for 
Polish, but want to use the Welsh voice to pronounce the phonemes?   
Unfortunately, Festival does not provide an easy way to separate the two 
phases.  Typically, it is done by modifying Festival itself by adding 
additional scheme code.  That's how the IMS German Festival mod works, for 
instance.  It is also possible to feed raw diphones into Festival, but again, 
you must write some scheme code.  Best I can say is to read the Festival 
manual ("info:festival in konqueror"), look at the festival code, and 
subscribe to the festival-talk mailing list.

You can probably make it all work with a shell script and some scheme code and 
use the Command plugin in KTTS temporarily until we figure out how to 
integrate your solution into KTTS.

You might also take a look at Hadifix, which already separates the processes 
into txt2pho and mbrola.

If you figure out a solution, please let me know.  I'd love to add Polish to 
KTTS's supported languages.

-- 
Gary Cramblitt (aka PhantomsDad)
KDE Text-to-Speech Maintainer
http://accessibility.kde.org/developer/kttsd/index.php