# Mockingjay analysis

## LinearLarge M3 (original)

| Layers | Phone | Speaker |
| :-----------: | :-----------: | :-----------: |
| WSN | 0.6513 | |
| 12 | 0.6487 | 0.9969 |
| 11 | 0.6551 | 0.9911 |
| 10 | 0.6322 | 0.9761 |
| 9 | 0.6291 | 0.8561 |
| 8 | 0.6181 | 0.632 |
| 7 | 0.601 | 0.91 |
| 6 | 0.5995 | 0.9215 |
| 5 | 0.6021 | 0.9478 |
| 4 | 0.6036 | 0.8001 |
| 3 | 0.6233 | 0.3099 |
| 2 | 0.6345 | 0.6357 |
| 1 | 0.5936 | 0.4168 |

## LinearLarge M3 (leo)

| Layers | Phone | Speaker |
| :-----------: | :-----------: | :-----------: |
| WSN | | 0.9550 |
| 12 | 0.6313 | 0.9751 |
| 11 | 0.6372 | 0.9839 |
| 10 | 0.6427 | 0.98 |
| 9 | 0.6412 | 0.9571 |
| 8 | 0.6382 | 0.9199 |
| 7 | 0.6355 | 0.848 |
| 6 | 0.6347 | 0.7187 |
| 5 | 0.6331 | 0.6033 |
| 4 | 0.6304 | 0.5597 |
| 3 | 0.6263 | 0.5653 |
| 2 | 0.621 | 0.5346 |
| 1 | 0.5969 | 0.4903 |

## LinearLarge M3 (pohan)

| Layers | Phone | Speaker |
| :-----------: | :-----------: | :-----------: |
| WSN | | 0.9904 |
| 12 | | 0.9023 |
| 11 | | 0.9458 |
| 10 | | 0.9877 |
| 9 | | 0.9893 |
| 8 | | 0.9913 |
| 7 | | 0.9907 |
| 6 | | 0.9533 |
| 5 | | 0.6367 |
| 4 | | 0.4094 |
| 3 | | 0.5367 |
| 2 | | 0.8426 |
| 1 | | 0.7114 |

## Phonme segmentation

| Model | R | F | Prec | Recall | Mean Product |
| :------------: | :-----: | :-----: | :-----: | :-----: | :----------: |
| Linearlargem3 | 0.79263 | 0.75552 | 0.74364 | 0.76779 | 1.60 |
| | 0.78469 | 0.74678 | 0.71205 | 0.78506 | 1.70 |
| | 0.77046 | 0.73720 | 0.67864 | 0.80682 | 1.80 |
| | 0.75100 | 0.72414 | 0.64790 | 0.83546 | 1.90 |
| | 0.72616 | 0.70832 | 0.61476 | 0.85020 | 2.00 |
| | 0.69921 | 0.69152 | 0.58276 | 0.86438 | 2.10 |
| | 0.79148 | 0.75863 | 0.77242 | 0.74533 | 1.50 |
| | 0.78279 | 0.75573 | 0.79460 | 0.72048 | 1.40 |
| | 0.77038 | 0.75157 | 0.82100 | 0.69296 | 1.30 |
| | 0.75518 | 0.74085 | 0.83223 | 0.66755 | 1.20 |
| | 0.73255 | 0.72193 | 0.84065 | 0.63259 | 1.10 |
| Linearlargem6 | 0.76209 | 0.72104 | 0.71964 | 0.72244 | 1.10 |
| | 0.75722 | 0.71438 | 0.67387 | 0.76007 | 1.12 |
| | 0.74891 | 0.70866 | 0.65183 | 0.77635 | 1.13 |
| | 0.70969 | 0.68307 | 0.59006 | 0.81089 | 1.15 |
| | 0.65632 | 0.52786 | 0.52786 | 0.84613 | 1.17 |
| | 0.55794 | 0.57122 | 0.42215 | 0.88305 | 1.20 |
| | 0.75851 | 0.72042 | 0.73648 | 0.70504 | 1.09 |
| | 0.75530 | 0.72211 | 0.75852 | 0.68903 | 1.08 |
| | 0.73960 | 0.71449 | 0.78660 | 0.65449 | 1.06 |
| | 0.72054 | 0.70284 | 0.81131 | 0.61996 | 1.04 |
| | 0.69050 | 0.67785 | 0.83040 | 0.57265 | 1.01 |
| Linearlargem9 | 0.74327 | 0.69807 | 0.69057 | 0.70574 | 1.20 |
| | 0.73540 | 0.68924 | 0.64060 | 0.74589 | 1.30 |
| | 0.70523 | 0.66625 | 0.58374 | 0.77593 | 1.40 |
| | 0.65948 | 0.63634 | 0.52351 | 0.81117 | 1.50 |
| | 0.60712 | 0.59777 | 0.46455 | 0.83812 | 1.60 |
| | 0.54253 | 0.54288 | 0.39576 | 0.86410 | 1.70 |
| | 0.73311 | 0.69301 | 0.71935 | 0.66853 | 1.10 |
| | 0.71446 | 0.68192 | 0.74842 | 0.62628 | 1.00 |
| | 0.68851 | 0.66000 | 0.76133 | 0.58247 | 0.90 |
| | 0.65990 | 0.63469 | 0.77762 | 0.53615 | 0.80 |
| | 0.62330 | 0.59572 | 0.78014 | 0.48181 | 0.70 |

| Model | R | F | Prec | Recall | Kernel Width |
| :------------: | :-----: | :-----: | :-----: | :-----: | :----------: |
| Baseline | 0.76649 | 0.72823 | 0.73747 | 0.71921 | 70 |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |

## Commands
- Linearlargem3

- Linearlargem6

python3 boundary/main.py --mode phoneme_segmentation --dataset timit --seg_target l --expname dryrun --dryrun --mock result/result_mockingjay/battle/public/linearlargem6/mockingjay-210000.ckpt  --headids 11-0 --logit_temp 1.1 --logit_h 0.08 --logit_blur 2.7

- Linearlargem9

python3 boundary/main.py --mode phoneme_segmentation --dataset timit --seg_target l --expname dryrun --mock result/result_mockingjay/battle/public/linearlargem9-2/mockingjay-260000.ckpt --headids 11-11 --logit_temp 1.5 --logit_blur 1.06  --dryrun

0.776858

- Baseline

python3 boundary/main.py --mode phoneme_segmentation --dataset timit --seg_target r --expname dryrun --dryrun --h 70 --K 6

- Attention KGM
python3 boundary/main.py --dataset timit --expname dryrun --mode phoneme_segmentation --seg_target lkgm --h 0.2 --window_size 32 --K 3 --logit_temp 1.05 --dryrun

0.7571119

- M9 11th:
python3 boundary/main.py --mode phoneme_segmentation --dataset timit --seg_target m --mock_layer 10 --expname dryrun --mock result/result_mockingjay/battle/public/linearlargem9-2/mockingjay-260000.ckpt --h 500 --logit_temp 1.004 --dryrun --plot_lines False --start 8 --end -8