[Accessibility] TTS API interface description + reqs update

[Gary Cramblitt]

On Wednesday 03 May 2006 05:21, Jonathan Duddington wrote:
> say_text()  End of text indication.
>
> 1.  It is not specified how end-of-text is indicated.  Perhaps a zero
> value character indicates end-of-text?
>
>
> 2.  Selection within the text.
> Consider an application which displays SSML text.  A user selects a
> portion of that text to be spoken.  How does the application indicate
> the end point of the text to be spoken?  The interface doesn't provide
> for an "end position".
>
> Does the application need to make a copy of the text, from the start of
> the <speak> block, up until the end point and then add a zero character
> terminator to indicate the end point?

Good point.

> Also, as I suggested earlier, it seems that the application needs to be
> able to specify the start point as a Character position (which is not
> currently provided) rather than a Word or Sentence position.

Hynek and I were discussing this last week.  We agree that one should be able 
to resume from a specified character position.  However, this raises two 
issues:

1.  What should the implementation do if the character position N is in the 
middle of an SSML tag?  I think it should resume from just after the tag 
(still applying the correct SSML context for the entire message).

2.  What should the implementation do if the character position N is in the 
middle of a word and we are not in spelling mode?  Speak from beginning of 
word?  End of word?  Spell the word fragment?  Attempt to speak the word 
fragment as best it can?  Leave up to the implementation/synthesizer?  Return 
an error?  Currently, AFAIK, Festival will spell the word fragment.  I don't 
think returning error is a good idea.

Thanks for the excellent feedback Jonathan.

-- 
Gary Cramblitt (aka PhantomsDad)