meeting-2008-08-22
Meeting on Sámi Synthetic speech
- Time: 080822
- Participants: Patrik Bye, Trond Trosterud, Martti Vainio
Presentation, background
Who are we and what do we do:
- hki
- divvun
- giellatekno
- patrik + intonation
Patrik
- segmental phon, noticed that no work on Sámi intonation done.
- 3-4 yrs project (meeting in September)
- Kickoff on basic empirical work in May
- Get hold of radio material, pedagogical tapes
- use the first year of the project for such things
- Two paid positions: Patrik + a phD.
- Applied to NFR to maybe get a third position as well.
- Autosegmental, metrical phonology, ToBi for Sámi
- Primarily North, also Lule and perhaps South.
Martti
- Done work on Finnish, Sámi would probably be closer to Finnsih
- ToBi for Finnish?
- We have one type of accent
- Important: Detaching form from function.
- Synthesis framework: Statistical intonation model.
- We leave it to the system to determine the contours.
- The system should learn the functions.
Collaborating?
- Patrik from Martti's synthesis?
- Martti from Patrik's phon analysis?
We would probably get information from each other, but we
Technically: Hki needs a good signal:
We have rule-based accentuation:
- 0 - copulas, function words
- 1 - verbs
- 2 - nouns, adjectives
- 3 - emphasis, like in narrow focus
No question intonation in Finnish.
- what about focus-triggering clitics? -- yes, utilized
- do you base yourself upon a syntactic analysis?) -- yes
- In addition, there is an analysis of the information structure (given - new)
Trond
We have the syntactic analysis + a forthcoming dependency analysis
- xerox two-level morphology + lexc
- vislcg3 disambiguation and dependency
"<Mu>" "mun" Pron Pers Sg1 Gen @>P #1->2 "<birra>" "birra" Po @ADVL #2->0 <========== #2->(5/4) "<son>" "son" Pron Pers Sg3 Nom @SUBJ> #3->4 "<lea>" "leat" <aux> V IV Ind Prs Sg3 @FS-STA #4->0 "<hupman>" "hupmat" <mv> V TV PrfPrc @ICL-AUX< #5->4 "<.>" "." CLB #6->0 "<Munge>" "mun" Pron Pers Sg1 Nom Foc/ge @SUBJ> #1->2 "<lean>" "leat" <aux> V IV Ind Prs Sg1 @FS-STA #2->0 "<boahtán>" "boahtit" <mv> V IV PrfPrc @ICL-AUX< #3->2 "<.>" "." CLB #4->0
Material
Radio news would be good, and studio programs.
Documents and background info
- Former application on tts project (Trond/Martti/Antti)
- Application/description of intonation project (Patrik)
- Report on Sámi tts to the Sámi Parliament (Sjur Moshagen 2003)
- Governmental whitepaper 2008 requiring Sámi tts (paragraph "Talesyntese" here):
Forthcoming milestones
Meeting between Divvun and the Sámi parliament in october
Upcoming deadlines?
Work planning, division of labour
Autumn 08:
- Exchange papers and background documents.
- Look for upcoming grant proposal deadlines
- Norway...
- Finland.. Martti to have a look
- Norway...
- Would it be possible to make a Sámi-as-Finnish before october?
- identify the Finnish text input format (Martti)
- make the same for Sámi (Patrik, Trond)
- syllabify the Sámi text
- turn the result into "Finnish orthography"
- syllabify the Sámi text
- Feed the result to the finnish tts and lean back.
- identify the Finnish text input format (Martti)
Spring 09 and onwards:
- Discuss formats for and evt. exchange empirical material
Resources
Needed
TODO:
- not a diphone system, but an hmm-based system.
- One man-year for Sámi segmental tts. ==> suggest 3 years (which includes
- getting to know the system and making the usual mistakes and false starts)
Find out how much to make tts
Available
- Martti: Finnish Academy researcher
- Antti: Funding for him until next summer
- Some grant proposals: ...
- Evt. hiring new persons for working on Sámi.
Q: Is it possible to integrate with standard OS's (Win/Mac) and common end-user applications (dyslectic, proofreading, screenreaders, etc.)?
The answer is yes. Web-based is easy. Integrating in mobile phones or text
Next meeting
Within a month.
Before that: the SPE rules and the Finnish example.