iFaz/whisper-SER-base-v4

first run with
iFaz/Whisper_Compatible_SER_benchmark
eavl: skip_special_token = True

rerun with this dataset
iFaz/Whisper_Compatible_SER_benchmark
eavl: skip_special_token = True

Evaluation result:

V4 (Rerun of Base-v1 --- Skip Special Token = True)

---

Dataset: test_CREMA_D Dataset

• Number of Samples: 2000
• Overall WA: 0.5720
• Overall UA: 0.5823
• Micro F1 Score: 0.5720
• Macro F1 Score: 0.5099

Classification Report:

              precision    recall  f1-score   support

           0       0.40      0.91      0.56       336
           1       0.76      0.09      0.16       346
           2       0.79      0.87      0.83       344
           3       0.71      0.88      0.79       288
           4       0.80      0.07      0.13       344
           5       0.53      0.67      0.59       342

    accuracy                           0.57      2000
   macro avg       0.67      0.58      0.51      2000
weighted avg       0.67      0.57      0.50      2000

Confusion Matrix:

[[306   0  14  10   2   4]
 [182  32  42  22   2  66]
 [ 36   2 300   6   0   0]
 [ 22   0   8 254   0   4]
 [170   6   8  10  24 126]
 [ 48   2   6  56   2 228]]

---

Dataset: test_ravdess Dataset

• Number of Samples: 1200
• Overall WA: 0.6300
• Overall UA: 0.6400
• Micro F1 Score: 0.6300
• Macro F1 Score: 0.6466

Classification Report:

              precision    recall  f1-score   support

           0       0.32      0.95      0.48       160
           1       1.00      0.33      0.49       160
           2       1.00      0.70      0.82       160
           3       0.79      0.59      0.67       400
           4       0.93      0.70      0.80       160
           5       0.66      0.57      0.61       160

    accuracy                           0.63      1200
   macro avg       0.78      0.64      0.65      1200
weighted avg       0.78      0.63      0.65      1200

Confusion Matrix:

[[152   0   0   8   0   0]
 [ 76  52   0  32   0   0]
 [ 28   0 112  20   0   0]
 [116   0   0 236   0  48]
 [ 44   0   0   4 112   0]
 [ 60   0   0   0   8  92]]

---

Dataset: test_IEMOCAP Dataset

• Number of Samples: 1000
• Overall WA: 0.4880
• Overall UA: 0.4542
• Micro F1 Score: 0.4880
• Macro F1 Score: 0.4715

Classification Report:

              precision    recall  f1-score   support

           0       0.55      0.51      0.53       370
           2       0.98      0.25      0.40       167
           3       0.36      0.70      0.48       269
           5       0.73      0.36      0.48       194

    accuracy                           0.49      1000
   macro avg       0.66      0.45      0.47      1000
weighted avg       0.61      0.49      0.48      1000

Confusion Matrix:

[[189   1 171   9]
 [ 39  42  83   3]
 [ 68   0 188  13]
 [ 47   0  78  69]]