RT07 12/5 telecon
11 AM EST

LDC: Stephanie Strassel, Meghan Glenn
IBM: Makis, Etienne, Jing Huang	
ICSI: Adam Janin, Wooters
UKA: Mathias, Cedric, John McDonough
LIMSI: Lori Lamel
AMI: Hain

-Review open action items
 -NIST starts a discussion on whether or not coffee breaks are 
  in-domain and therefore eligible to be in the testset.
 -CHIL to release coffee break examples
  --> Cedric did make these available in dev data, and they're clearly 
marked.  They are not available as a /separate/ dataset, but they are 
clearly separated from the rest of the dev data in the release.  Cedric 
points out that not many sites asked for this data. Adam @ ICSI did 
request it, but wasn't able to process it -- or  even read the disk on 
any OS.
 **Cedric will give access to download to interested parties.
 **Everyone will review coffee break examples and comment next time.

 -All review examples
 **postponed until next telecon

 -NIST publishes the Forced Aligned reference files from RT-05 and RT-06
 No comment.

 -CHIL will report on the soon-to-be-released DEV data and it will be 
  documented on a website.

  Done

 -Adam J. to release "quick-and-dirty" scripts for processing AMI 
  transcripts
 ** will circulate

 -Adam J. to circulate suggested AMI Dev/Train divide
 ** Will just pick one and circulate

 -Speaker diarization of lecture data: JF and SB report.
  This year's lecture data is much more interactive than last year's
  data: enough to meeth the 10% threshhold so the lecture data will be
  used for the SPKR task

--> I GOT CUT OFF FOR THE REST OF THIS

 -Pending email discussion of close talking mic condition: "Can manual 
segmentation be a primary condition for the IHM tests?"
 Yes. Adam says manual segmentation is okay. ICSI will still do 
segmentation, but want the manual reference to see how much is lost in 
automatic segmentation.
 Lots of cross-talk. LL raises concern about this point. Sites agree 
that cross-talk and overlapping speech are good -- this is one 
motivation for making manual segmentation a primary condition. JF 
suggests conditionally scoring everything and then sub-scoring on 
segments with or without cross-talk.

 - Pending email discussion to keep MDM condition the primary condition
   All agreed: should be primary. But JF will start email discussion to 
   confirm with others.

-New Issues:

 -SASTT task definition
  In eval plan. Not many people have had a chance to review that 
  document yet. **Let's discuss this at the next meeting, and over 
  email.

 -Evaluation Schedule discussion
  - no complaints so it will stand


 -Evaluation data
   -CMU delivered data today
    No comment.

 -Training data:

CHIL: Description: 40 segments of 5 min from all over seminar scenario 
-- some coffee breaks, some beginning of seminar, some end of seminar, 
some "normal" data as have had in the past. This data looks exactly like 
the test data, so if sites do not like the coffee breaks, they should 
speak up now, and then there will only be ~35 segments in the test 
set. 

NIST: will be releasing training data before end of the year: full 
training set is 18 hours, will distribute as much as is complete by 
then. (at least 10 hours) If sites want this data, send NIST an email.

Adam points out that the AMI meetings are not on the list of training 
meetings.

**ACTION: JF will add these to the list.

 - NIST has secured a limited amount of funding for the release of LDC 
corpora to RT participants. The list of corpora are available in the 
eval plan (and also in Stephanie's email of 12/5). Stephanie explains 
the data-access process: all who are interested in the LDC data must 
send an email to NIST, requesting it. NIST will send LDC the list; LDC 
will circulate user agreements to sites. This is a no-cost membership 
for the duration of the evaluation, and sites can request any corpora on 
the lists, which includes FISHER, TDT4, and RT corpora from previous 
years. JF underscores that there are LIMITED funds for this purpose, so 
sites are asked to please only request what they NEED. Stephanie 
concludes that LDC will be able to accomodate any reasonable requests.

**ACTION: Sites send NIST email if they want this data.

-Schedule next telecon

 12/21, 11 AM EST.

Call adjourned 11:49 AM