[mary-users] Accent vs. Stress

[Ingmar Steiner]

For the record, http://mary.opendfki.de/ticket/269, fixed in http://mary.opendfki.de/changeset/2375 (branch) and http://mary.opendfki.de/changeset/2378 (trunk).

Due to an oversight, this bug was not assigned a milestone and therefore does not show up in the 4.1.0 release notes... =(

Best wishes,

/**
 * Ingmar Steiner
 * Researcher, Language Technology
 * German Research Center for Artificial Intelligence
 *
 * Campus D3 1 +1.18
 * D-66123 Saarbrücken
 * Germany
 * Phone: ++49-681-857-75-5263 (NEW!)
 * Email: ingmar.steiner at dfki.de
 *
 * Deutsches Forschungszentrum für Künstliche Intelligenz GmbH
 * Trippstadter Straße 122, D-67663 Kaiserslautern, Germany
 * Geschäftsführung:
 * Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)
 * Dr. Walter Olthoff
 * Vorsitzender des Aufsichtsrats:
 * Prof. Dr. h.c. Hans A. Aukes
 * Amtsgericht Kaiserslautern, HRB 2313
 */

On 12 Oct 2010, at 16:07, Thorsten Westermann wrote:

> Hello again!
> 
> Ingmar, could you please tell what and where this "certain bug fix" is?
> I don't find it in the release notes of 4.1.0
> 
> Thank you!
> 
> MfG,
> Westermann
> 
> -------- Original-Nachricht --------
>> Datum: Tue, 12 Oct 2010 10:42:43 +0200
>> Von: Ingmar Steiner <ingmar.steiner at dfki.de>
>> An: Thorsten Westermann <thorstenwestermann at gmx.net>
>> CC: mary-users <mary-users at dfki.de>
>> Betreff: Re: [mary-users] Accent vs. Stress
> 
>> Dear Thorsten,
>> 
>> just to put my $0.02 in, in the context of how MARY uses stress and
>> accent, the following may grant additional insight:
>> 
>> Stress is determined at the lexical level (in PHONEMES), and works as a
>> ternary attribute of syllables; every syllable can have (a) primary, (b)
>> secondary, or (c) no stress, with no more than two stresses per word. Since
>> stressed syllables are much less likely to be reduced (i.e. shortened, with
>> centralized vowels) this plays an important role in determining which units
>> to select during synthesis.
>> 
>> Stressed syllables are also the "anchors" for accents, which are predicted
>> in INTONATION; accents cannot normally fall on unstressed syllables. In
>> Mary, accents are described by ToBI labels ("L+H*" and so on) and refer
>> mainly to F0 targets and contours. So stress influences accent (and this is why
>> a certain bugfix in 4.1 should have improved these things considerably),
>> and both are important symbolic features for acoustic parameter prediction
>> and unit selection (both directly, and indirectly, through acoustically
>> predicted continuous features).
>> 
>> Best wishes,
>> 
>> /**
>> * Ingmar Steiner
>> * Researcher, Language Technology
>> * German Research Center for Artificial Intelligence
>> *
>> * Campus D3 1 +1.18
>> * D-66123 Saarbrücken
>> * Germany
>> * Phone: ++49-681-857-75-5263 (NEW!)
>> * Email: ingmar.steiner at dfki.de
>> *
>> * Deutsches Forschungszentrum für Künstliche Intelligenz GmbH
>> * Trippstadter Straße 122, D-67663 Kaiserslautern, Germany
>> * Geschäftsführung:
>> * Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)
>> * Dr. Walter Olthoff
>> * Vorsitzender des Aufsichtsrats:
>> * Prof. Dr. h.c. Hans A. Aukes
>> * Amtsgericht Kaiserslautern, HRB 2313
>> */
>> 
>> On 11 Oct 2010, at 11:56, Thorsten Westermann wrote:
>> 
>>> Hi Nickolay,
>>> 
>>> okay, I think I have understood it:
>>> 
>>> 1)
>>> The stress is predefined, it's always the same. Or it should be, as long
>> as the person doesn't speak a dialect or so.
>>> 
>>> 2)
>>> But the accent can be totally different. 
>>> Example about accents for the word "eyes":
>>> 
>>> "You got EYES!" (="See for yourself, don't ask me!")
>>> "You got EYES!" (=Wonderful eyes!)
>>> "You GOT eyes!" (=You are not blind!)
>>> "You got eyes..." (=They are still there, stop bothering)
>>> Something like that.
>>> 
>>> Did I understand correctly?
>>> 
>>> When this is correct, how can Mary make assumptions about how a word is
>> accented? I see this byte feature, e. g. "Accented_Syls_From_Phrase_Start".
>>> 
>>> Does Mary look at the pitch or loudness or similiar, or is it a
>> rule-based assumption that a unit is accented and in which way?
>>> 
>>> MfG,
>>> Westermann
>>> 
>>> -------- Original-Nachricht --------
>>>> Datum: Mon, 11 Oct 2010 09:37:28 +0400
>>>> Von: "Nickolay V. Shmyrev" <nshmyrev at nexiwave.com>
>>>> An: Thorsten Westermann <thorstenwestermann at gmx.net>
>>>> CC: Christopher Bader <cb at kratylos.com>, mary-users at dfki.de
>>>> Betreff: Re: [mary-users] Accent vs. Stress
>>> 
>>>> 2010/10/11, Thorsten Westermann <thorstenwestermann at gmx.net>:
>>>>> Hello!
>>>>> 
>>>>> Thanks for your reply.
>>>>> I was wondering why we have both stress and accent features when they
>>>> are
>>>>> basically the same, and just didn't come to any conclusion.
>>>>> 
>>>> 
>>>> They are not the same. Traditionally in Festival "stress" is a lexical
>>>> stress, something that is specified by lexicon. Accent is intonational
>>>> feature, it depends on phrase. For example accent could be present
>>>> only in content words. Accent could be derived from presence of
>>>> intonation events for example from ToBI event in the syllable. Or it
>>>> can be predicted independently and be the feature for ToBI event
>>>> prediction.
>>> 
>>> -- 
>>> GRATIS! Movie-FLAT mit über 300 Videos. 
>>> Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome
>>> _______________________________________________
>>> Mary-users mailing list
>>> Mary-users at dfki.de
>>> http://www.dfki.de/mailman/cgi-bin/listinfo/mary-users
>> 
> 
> -- 
> GRATIS! Movie-FLAT mit über 300 Videos. 
> Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome