Speech Emotion Recognition 연구 기록

맨날 공책에 연구기록 정리해놔도 다 없어진다. 그래서 이제 웹에 저장해볼까 한다.

1. CRNN, layer = [2, 2, 3], filters = [64, 128, 256]

epoch 19 loss 0.6085 acc 0.8649
epoch 41 loss 0.6767 acc 0.8198
epoch 15 loss 0.5758 acc 0.8378
epoch 37 loss 0.6570 acc 0.8108

2. CRNN, layer = [2, 3, 3], filters = [64, 128, 256]

epoch 10 loss 0.8609 acc 0.7297
epoch 7 loss 0.6213 acc 0.7838
epoch 41 loss 0.8283 acc 0.8378
epoch 35 loss 0.6847 acc 0.8468

3. CRNN, layer = [3, 3, 3], filters = [64, 128, 256]

epoch 28 loss 0.6859 acc 0.8198
epoch 33 loss 0.7510 acc 0.8198
epoch 17 loss 0.7368 acc 0.8288
epoch 17 loss 1.2993 acc 0.6757

4. CRNN, layer = [2, 2, 2], filters = [64, 128, 256]

epoch 16 loss 0.6999 acc 0.8108
epoch 18 loss 0.5953 acc 0.8198
epoch 8 loss 0.7030 acc 0.7568
epoch 11 loss 0.6641 acc 0.8288

5. resnetRNN, layer = [5, 4, 4], filters = [64, 128, 256]

epoch 4 loss 0.7461 acc 0.7568
epoch 9 loss 0.9223 acc 0.6937

6. resnetRNN, layer = [3, 2, 2], filters = [64, 128, 256]

epoch 4 loss 0.9112 acc 0.7027
epoch 7 loss 0.7967 acc 0.7117

7. resnetRNN, layer = [7, 6, 6], filters = [64, 128, 256]

epoch 13 loss 1.5899 acc 0.6757
epoch 9 loss 0.9989 acc 0.7027

8. resnetRNN, layer = [9, 8, 8], filters = [64, 128, 256]

epoch 18 loss 1.2800 acc 0.6667
epoch 4 loss 1.0925 acc 0.5856

9. resnetRNN, layer = [3, 4, 4], filters = [64, 128, 256]

epoch 14 loss 0.9747 acc 0.6757
epoch 14 loss 0.8728 acc 0.7748
epoch 10 loss 0.8545 acc 0.7297
epoch 5 loss 0.8566 acc 0.6847

10. resnetRNN, layer = [3, 4, 4], filters = [64, 128, 256], non_initialization

epoch 14 loss 0.8709 acc 0.7658
epoch 9 loss 0.9039 acc 0.7568

11. resnetRNN, layer = [5, 4, 4], filters = [64, 128, 256], non_initialization

epoch 14 loss 0.9539 acc 0.7658
epoch 13 loss 0.9006 acc 0.7748

12. resnetRNN, layer = [7, 6, 6], filters = [64, 128, 256], non_initialization

epoch 11 loss 1.2741 acc 0.5946
epoch 19 loss 1.5301 acc 0.6577

13. resnetRNN, layer = [3, 2, 2], filters = [64, 128, 256], non_initialization

epoch 27 loss 1.6902 acc 0.6306
epoch 12 loss 1.6095 acc 0.6667

14. CRNN, layer = [3, 2, 2], filters = [64, 128, 256]

epoch 20 loss 1.1918 acc 0.7568
epoch 18 loss 1.2211 acc 0.7568

15. CRNN, layer = [3, 4, 4], filters = [64, 128, 256]

Epoch 16 (Test) Loss 0.9358 Acc 0.7477
Epoch 16 (Test) Loss 1.0985 Acc 0.7838
Epoch 28 (Test) Loss 1.1612 Acc 0.7477
Epoch 20 (Test) Loss 1.0648 Acc 0.7568

16. googlenetRNN, module = [1, 1], filters = [64, 128]

Epoch 16 (Test) Loss 0.9348 Acc 0.7944
Epoch 16 (Test) Loss 0.8571 Acc 0.7757
Epoch 13 (Test) Loss 1.0103 Acc 0.7570
Epoch 14 (Test) Loss 0.6934 Acc 0.8037

17. googlenetRNN, module = [1, 1, 1], filters = [64, 128, 128]

Epoch 22 (Test) Loss 0.8625 Acc 0.8224
Epoch 9 (Test) Loss 0.8106 Acc 0.7757
Epoch 8 (Test) Loss 0.6882 Acc 0.7944
Epoch 8 (Test) Loss 0.7558 Acc 0.8037

18. googlenetRNN, module = [1, 2, 1], filters = [64, 128, 128]

Epoch 5 (Test) Loss 0.8984 Acc 0.7196
Epoch 17 (Test) Loss 0.7758 Acc 0.8037
Epoch 7 (Test) Loss 1.0300 Acc 0.7196
Epoch 11 (Test) Loss 0.8914 Acc 0.6916

19. googlenetRNN, module = [1, 1, 1, 1], filters = [64, 128, 128], maxpooling * 3

Epoch 9 (Test) Loss 0.8737 Acc 0.7944
Epoch 8 (Test) Loss 0.6057 Acc 0.7757
Epoch 10 (Test) Loss 0.8225 Acc 0.7570
Epoch 4 (Test) Loss 0.9024 Acc 0.7664

20. googlenetRNN, module = [1, 2, 1, 1], filters = [64, 128, 128], maxpooling * 3

Epoch 11 (Test) Loss 0.8609 Acc 0.7383
Epoch 20 (Test) Loss 0.9937 Acc 0.7944
Epoch 4 (Test) Loss 0.8394 Acc 0.7103
Epoch 11 (Test) Loss 0.7857 Acc 0.7477