wav2vec2-base-960h-paper

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the HTS98/ORIGINAL_VER1.2 - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.8214
• Wer: 0.9640

Model description

More information needed

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 16
• 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
• lr_scheduler_warmup_steps: 420
• num_epochs: 50.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.99	104	5.6840	1.0
No log	2.0	209	3.9772	1.0
No log	3.0	314	3.4204	1.0
No log	4.0	419	3.3692	1.0
5.5612	4.99	523	3.3945	1.0
5.5612	6.0	628	3.3426	1.0
5.5612	7.0	733	3.3333	1.0
5.5612	8.0	838	3.3296	1.0001
5.5612	8.99	942	3.1853	0.9999
3.2743	10.0	1047	2.1381	1.0245
3.2743	11.0	1152	1.6965	1.0142
3.2743	12.0	1257	1.4230	1.0011
3.2743	12.99	1361	1.2679	0.9873
3.2743	14.0	1466	1.1570	0.9836
1.5432	15.0	1571	1.0858	0.9784
1.5432	16.0	1676	1.0303	0.9769
1.5432	16.99	1780	0.9855	0.9746
1.5432	18.0	1885	0.9559	0.9709
1.5432	19.0	1990	0.9328	0.9728
0.902	20.0	2095	0.9166	0.9738
0.902	20.99	2199	0.8991	0.9698
0.902	22.0	2304	0.8717	0.9681
0.902	23.0	2409	0.8665	0.9669
0.7003	24.0	2514	0.8589	0.9670
0.7003	24.99	2618	0.8420	0.9659
0.7003	26.0	2723	0.8473	0.9661
0.7003	27.0	2828	0.8543	0.9666
0.7003	28.0	2933	0.8315	0.9623
0.5914	28.99	3037	0.8281	0.9626
0.5914	30.0	3142	0.8315	0.9625
0.5914	31.0	3247	0.8261	0.9620
0.5914	32.0	3352	0.8214	0.9640
0.5914	32.99	3456	0.8310	0.9634
0.5157	34.0	3561	0.8252	0.9635
0.5157	35.0	3666	0.8373	0.9638
0.5157	36.0	3771	0.8422	0.9629
0.5157	36.99	3875	0.8294	0.9632
0.5157	38.0	3980	0.8332	0.9576
0.4655	39.0	4085	0.8330	0.9595
0.4655	40.0	4190	0.8297	0.9625
0.4655	40.99	4294	0.8365	0.9621
0.4655	42.0	4399	0.8361	0.9621
0.4266	43.0	4504	0.8416	0.9625
0.4266	44.0	4609	0.8381	0.9634
0.4266	44.99	4713	0.8448	0.9645
0.4266	46.0	4818	0.8447	0.9625
0.4266	47.0	4923	0.8464	0.9641
0.4019	48.0	5028	0.8449	0.9628
0.4019	48.99	5132	0.8487	0.9626
0.4019	49.64	5200	0.8465	0.9629

Framework versions

• Transformers 4.31.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.7.0
• Tokenizers 0.13.3