wav2vec2-large-robust-paper

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

• Loss: 0.8696
• Wer: 0.4572

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: 10
• eval_batch_size: 8
• seed: 42
• 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	1.0	335	3.9163	1.0
7.1369	2.0	670	3.3422	1.0
3.3448	3.0	1005	3.3355	1.0
3.3448	4.0	1340	3.3263	1.0
3.3277	5.0	1675	2.8928	1.0079
2.6655	6.0	2010	1.7822	0.8788
2.6655	7.0	2345	1.3193	0.7055
1.4617	8.0	2680	1.1408	0.6070
1.0805	9.0	3015	1.0108	0.5422
1.0805	10.0	3350	0.9517	0.5154
0.8759	11.0	3685	0.9082	0.4902
0.7462	12.0	4020	0.8758	0.4706
0.7462	13.0	4355	0.8696	0.4572
0.6429	14.0	4690	0.8731	0.4535
0.5672	15.0	5025	0.8749	0.4508
0.5672	16.0	5360	0.8753	0.4512
0.4959	17.0	5695	0.9039	0.4487
0.4456	18.0	6030	0.9161	0.4433
0.4456	19.0	6365	0.9506	0.4430
0.392	20.0	6700	0.9412	0.4439
0.3594	21.0	7035	0.9884	0.4416
0.3594	22.0	7370	1.0222	0.4510
0.3175	23.0	7705	1.0345	0.4439
0.2947	24.0	8040	1.0849	0.4465
0.2947	25.0	8375	1.0879	0.4472
0.2674	26.0	8710	1.1071	0.4512
0.2521	27.0	9045	1.1147	0.4494
0.2521	28.0	9380	1.1426	0.4525
0.2321	29.0	9715	1.1592	0.4440
0.2235	30.0	10050	1.1782	0.4450
0.2235	31.0	10385	1.2050	0.4437
0.2071	32.0	10720	1.2224	0.4400
0.1951	33.0	11055	1.2270	0.4471
0.1951	34.0	11390	1.2466	0.4483
0.1892	35.0	11725	1.2325	0.4429
0.1809	36.0	12060	1.2755	0.4427
0.1809	37.0	12395	1.2675	0.4422
0.1746	38.0	12730	1.3022	0.4418
0.1656	39.0	13065	1.3179	0.4408
0.1656	40.0	13400	1.2934	0.4425
0.1614	41.0	13735	1.3304	0.4426
0.1564	42.0	14070	1.3148	0.4420
0.1564	43.0	14405	1.3267	0.4433
0.1546	44.0	14740	1.3331	0.4413
0.1515	45.0	15075	1.3445	0.4388
0.1515	46.0	15410	1.3530	0.4372
0.147	47.0	15745	1.3443	0.4385
0.1447	48.0	16080	1.3503	0.4369
0.1447	49.0	16415	1.3590	0.4393
0.1437	50.0	16750	1.3668	0.4372

Framework versions

• Transformers 4.31.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.7.0
• Tokenizers 0.13.2