Bagus/wav2vec2_swbd_emodb

finetuned

This model is a fine-tuned version of facebook/wav2vec2-large-robust-ft-swbd-300h on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.0281
• Uar: 0.7318
• Acc: 0.7721

For the test set:

• UAR: 0.74
• ACC: 0.794

Model description

This model is to predict four emotion categories given and audio file. Labels are anger', 'happiness', 'sadness', 'neutral'. This wav2vec2-based model is known cannot detect 'happiness'.

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Uar	Acc
No log	0.15	1	1.3899	0.25	0.1985
No log	0.31	2	1.3850	0.25	0.1985
No log	0.46	3	1.3815	0.25	0.1985
No log	0.62	4	1.3772	0.25	0.1985
No log	0.77	5	1.3714	0.25	0.4044
No log	0.92	6	1.3656	0.25	0.4044
1.4878	1.08	7	1.3610	0.25	0.4044
1.4878	1.23	8	1.3583	0.25	0.4044
1.4878	1.38	9	1.3549	0.25	0.4044
1.4878	1.54	10	1.3518	0.25	0.4044
1.4878	1.69	11	1.3491	0.25	0.4044
1.4878	1.85	12	1.3458	0.25	0.4044
1.4878	2.0	13	1.3425	0.25	0.4044
1.2316	2.15	14	1.3401	0.25	0.4044
1.2316	2.31	15	1.3380	0.25	0.4044
1.2316	2.46	16	1.3354	0.25	0.4044
1.2316	2.62	17	1.3326	0.25	0.4044
1.2316	2.77	18	1.3292	0.2778	0.4265
1.2316	2.92	19	1.3250	0.2963	0.4412
1.3835	3.08	20	1.3212	0.3519	0.4853
1.3835	3.23	21	1.3158	0.4029	0.5221
1.3835	3.38	22	1.3096	0.5047	0.6029
1.3835	3.54	23	1.3019	0.5695	0.6544
1.3835	3.69	24	1.2944	0.6485	0.7059
1.3835	3.85	25	1.2856	0.6534	0.6985
1.3835	4.0	26	1.2773	0.6768	0.7059
1.1038	4.15	27	1.2688	0.6540	0.6691
1.1038	4.31	28	1.2554	0.6404	0.6471
1.1038	4.46	29	1.2404	0.6359	0.6397
1.1038	4.62	30	1.2222	0.6586	0.6765
1.1038	4.77	31	1.2057	0.6631	0.6838
1.1038	4.92	32	1.1874	0.6769	0.6985
1.075	5.08	33	1.1624	0.6953	0.7206
1.075	5.23	34	1.1427	0.7182	0.75
1.075	5.38	35	1.1270	0.7182	0.75
1.075	5.54	36	1.1085	0.7227	0.7574
1.075	5.69	37	1.0982	0.7227	0.7574
1.075	5.85	38	1.0943	0.7227	0.7574
1.075	6.0	39	1.0930	0.7136	0.7426
0.7211	6.15	40	1.0903	0.7091	0.7353
0.7211	6.31	41	1.0858	0.7091	0.7353
0.7211	6.46	42	1.0816	0.7045	0.7279
0.7211	6.62	43	1.0734	0.7091	0.7353
0.7211	6.77	44	1.0617	0.7136	0.7426
0.7211	6.92	45	1.0536	0.7136	0.7426
0.6595	7.08	46	1.0450	0.7318	0.7721
0.6595	7.23	47	1.0370	0.7364	0.7794
0.6595	7.38	48	1.0323	0.7364	0.7794
0.6595	7.54	49	1.0301	0.7364	0.7794
0.6595	7.69	50	1.0307	0.7364	0.7794
0.6595	7.85	51	1.0302	0.7318	0.7721
0.6595	8.0	52	1.0307	0.7318	0.7721
0.5067	8.15	53	1.0317	0.7318	0.7721
0.5067	8.31	54	1.0324	0.7318	0.7721
0.5067	8.46	55	1.0324	0.7318	0.7721
0.5067	8.62	56	1.0326	0.7273	0.7647
0.5067	8.77	57	1.0315	0.7318	0.7721
0.5067	8.92	58	1.0297	0.7318	0.7721
0.5617	9.08	59	1.0287	0.7318	0.7721
0.5617	9.23	60	1.0281	0.7318	0.7721

Framework versions

• Transformers 4.32.0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.13.3