[Speechd] KTTS and Sentence Boundary Detection

[Milan Zamazal]

>>>>> "GC" == Gary Cramblitt <address@hidden> writes:

    GC> In practice, so far, only the Polish language has needed a
    GC> separate SBD filter, mostly because Polish Festival incorrectly
    GC> "speaks" punctuation characters. 

Shouldn't Polish Festival be fixed instead?

    GC> 1) we don't handle SSML "relative" attributes (<prosody
    GC> rate="+10">), 2) Festival seems to have trouble with voice
    GC> attributes, so we strip them out when converting to SABLE, and

We support both relative prosody values and voice switching in the
festival-freebsoft-utils SSML processor.

    GC> 3) we don't handle the <say-as> tag, which isn't fully defined
    GC> in the SSML spec anyway.

Neither we do, except for our own special attribute value
detail="spell".

    GC> The ideal synth would not perform SBD on the entire input before
    GC> it begins speaking the first sentence, but I don't believe
    GC> Festival is so optimized.  Does Speech Dispatcher do something
    GC> to solve this problem?  

We use festival-freebsoft-utils for this purpose.  In case of plain
text, Festival processes only as much of text as is needed for creating
a complete utterance.  In case of SSML, the document has to be first
parsed with a SSML parser, but this is usually fast and definitely much
faster than complete text processing.  So we have no delays because of
utterance chunking in large documents.

See http://www.freebsoft.org/festival-freebsoft-utils for more
information.

Regards,

Milan Zamazal

-- 
Life.  Don't talk to me about life.              -- Marvin the Paranoid Android