SPINE2

SPeech in Noisy Environments - Phase 2

Data Preparation Procedures


ARCON Collects
 
Data Talkers Pairs Coders Noises Sessions Minutes Hours
TRAIN 6 3 4 8 96 288 4.8
DEV/TEST 4 2 4 8 64 192 3.2
EVAL 32 16 4 8 512 1536 25.6

For a total of 42 talkers, 672 sessions, 33.6 hours of data.

Arcon will deliver all of this data to LDC at the rate of ~4 talker pairs per week.  LDC will identify, format and transcribe the subset of data
that has been specified to support this year's evaluation:

LDC Prepares
 
Data Talkers Pairs Coders Noises Sessions Minutes Hours
TRAIN 4 2 4 8 64 192 3.2
DEV/TEST 4 2 4 4 32 96 1.6
EVAL 32 16 4 8 64** 192 3.2
(**The eval data will be subsampled; not all combinations will be used so the total number of sessions amounts to 64, not 512.)
 


strassel@ldc.upenn.edu

7/18/2001