Meeting Recording Pool for RT-05S the Evaluation

Updated: February 11, 2005

The following table summarizes the meetings that will be used to construct the RT-05S evaluation corpus.  They will be contributed by five research organizations.  CMU, ICSI and NIST will contribute meetings to the '05 evaluation that are similar in nature to the meetings used in the '04 evaluation.  The AMI and CHIL programs are new data contributors to the evaluation.   The AMI program will contribute conference room meetings, while the CHIL program will contribute lecture room meetings. 

The table below describes the meetings contributed by each organization.  This information, some of which is not legal side information for a system, is intended not to describe the makeup of the evaluation corpus, but rather a complete accounting of all the sensors collected for each meeting.  Some of the sensors may not be published in the evaluation corpus.


Data Contributors
AMI
CMU
ICSI
NIST CHIL
Mtg Info Meeting type
Conference
Conference
Conference
Conference
Lecture
# of meetings 2-3 2 2  2 ?
Duration
? 20 minutes 1 hour  30-60 minutes
?
# of participants per mtg (Not permissible side information)
4 4 7-9  ? 1 lecturer
small number of students
Sensor Info Head Mic Yes
Yes Yes  Yes
Yes
(Lecture and 3-4 students)
Lapel Mic
Yes
No
No
Yes
No
Central distance mic 1 centrally placed circular array of 8 mics
3 omni-directional table mic

4 omni directional mics down the center

2 electrets mounted on mock PDA

3 omni directional mics placed down the center of the table

4-way directional microphone
Some table top mics
Source localization array
 (CHIL's upside down "T" array)
No No No  No 4 arrays
Mark III array No Yes (64-channels) No  No 1 array
Camera

1 close-up per participant

2 room-view

2 camcorders No 7 Hi Def digital cameras, one is a pan/tilt

Mpeg-2 output
4 640x480 video views, one is a pan/tilt/zoom
Annotation

Segmented by utterance

Included speaker ID and gender

None

Time-marked speaker turns

Vocal/non-vocal noises

Notes on weird pronunciations

Info about microphones and speakers

Time-marked  speaker turns

Transcribed
Time-marked speaker turns
Transcribed

Source localization annotations
Sample Rate 16kHz. 16-bit 16kHz, 16-bit 16kHz, 16-bit  Native 48 Khz, 24 bit
44.1Khz, 24 bit
Training Data
N/A
LDC Catalog: LDC2004S05
LDC2004T10
LDC Catalog: LDC2004S02 LDC2004T04 LDC Catalog: LDC2004S09 LDC2004T13 N/A
Development Test Data
A devtest will be made available
RT-04S Eval Set
RT-04S Eval Set
RT-04S Eval Set
5 Meetings soon to be released
URL AMI Meeting Room Data Collection Effort   ICSI Meeting Room Data Collection Effort NIST Meeting Room Project