[mary-users] Controlling intonation in MARY

[Marc Schröder]

Hi Jörg,

(let's hold this type of discussions on the mary-users mailing list -- 
others might have a similar question, now or in the future, and the 
mary-users archive at http://www.dfki.de/pipermail/mary-users/ will 
allow people to browse older emails).

You want to create German poems, using MARY to convert GToBI-annotated 
text into concrete F0 values. As a consequence, you require the 
possibility to accent every other syllable of a phrase.

The prosody concept in MARY normally envisages each word (or token, to 
be more precise) to have zero or one accent. The TobiContourGenerator 
code therefore goes through all the *tokens* in a phrase that have an 
accent, determines the stressed syllable of that token, and attaches the 
F0-time targets for the token's accent to that syllable:
http://mary.opendfki.de/browser/trunk/java/de/dfki/lt/mary/modules/TobiContourGenerator.java#L713

Now, you could change that quite easily. As you can see when you look at 
the input format to the TobiContourGenerator (DURATIONS_EN or 
DURATIONS_DE), the syllables already have an "accent" attribute, which 
is currently ignored by the TobiContourGenerator. Instead of iterating 
over all *tokens* in the TobiContourGenerator, you could thus iterate 
over all *syllables* and test whether they are accented. Skip the 
getStressedSyllable() part, you already have your reference syllable... 
and you should get the realisation of an accent on every syllable that 
carries an accent attribute.

 From what I see, your change would be non-breaking for existing code, 
so feel free to submit this as a ticket on the development page, and 
contribute your code changes!

Best,

Marc


Jörg Bröggelwirth schrieb:
> Hallo Marc,
> 
> wie ich kuerzlich schon einmal schrieb, moechte ich ein GToBI Modell 
> fuer deutsche Gedichte mit jeweils festem Metrum erstellen. Dann wuerde 
> ich gerne Mary benutzen, um aus dem gegebenen GToBI Input konkrete F0 
> Werte zu berechnen. Im Extremfall kann es bei den Gedichten vorkommen, 
> dass jede zweite Silbe eines Verses, respektive Phrase, einen GToBI 
> Akzent zugewiesen bekommt. Meine Frage nun ist, ob eine so hohe Dichte 
> an Pitchakzenten von Mary ueberhaupt akzeptiert wird oder ob 
> standardmaessig einige Akzente getilgt wuerden. Aus den relevanten 
> Veroeffentlichungen zu Mary ging das aus meiner Sicht nicht klar hervor.
> 
> Viele Gruesse,
>    Joerg Broeggelwirth
> 

-- 
Dr. Marc Schröder, Senior Researcher
DFKI GmbH, Stuhlsatzenhausweg 3, D-66123 Saarbrücken, Germany
http://www.dfki.de/~schroed
Here. Now. Real, first-person experience. Am I there to witness it?