model's performance metrics
#3
by FENGYinLeo - opened
Could you please provide the model's performance metrics for dimensional affect prediction, including accuracy and the Concordance Correlation Coefficient (CCC)? I did not find these figures disclosed in the paper.
Hi, we have not evaluated this model on SG eval datasets since they don't have official dimensional affect scores. Regarding MSP 1.11-based eval sets, we internally evaluated it first to check consistency since our main goal was to improve classification accuracy. MSP results were: CCC for Valence: 0.6164
CCC for Arousal: 0.6595
CCC for Dominance: 0.5933, for IEMOCAP: CCC for Valence: 0.5344
CCC for Arousal: 0.7006
CCC for Dominance: 0.5305. Note that both datasets were not part of training due to licence
thanks.