wav2vec2-base-librispeech-model

This model is a fine-tuned version of facebook/wav2vec2-base on the LIBRI10H - ENG dataset. It achieves the following results on the evaluation set:

• Loss: 0.5515
• Wer: 0.4641

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: 0.0003
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.544	1.4493	500	1.7568	0.9820
1.42	2.8986	1000	1.0275	0.8168
1.0403	4.3478	1500	0.8305	0.7173
0.8574	5.7971	2000	0.7293	0.6649
0.7315	7.2464	2500	0.6632	0.6025
0.6389	8.6957	3000	0.6286	0.5695
0.5679	10.1449	3500	0.6102	0.5489
0.5085	11.5942	4000	0.5863	0.5215
0.4579	13.0435	4500	0.5661	0.4933
0.4097	14.4928	5000	0.5646	0.4823
0.382	15.9420	5500	0.5515	0.4644
0.3426	17.3913	6000	0.5585	0.4514
0.32	18.8406	6500	0.5598	0.4475
0.2926	20.2899	7000	0.6246	0.4587
0.2735	21.7391	7500	0.5887	0.4439
0.257	23.1884	8000	0.5978	0.4355
0.2409	24.6377	8500	0.5721	0.4215
0.2246	26.0870	9000	0.5979	0.4187
0.2139	27.5362	9500	0.6103	0.4145
0.2014	28.9855	10000	0.6436	0.4157
0.1917	30.4348	10500	0.6471	0.4188
0.184	31.8841	11000	0.6410	0.4068
0.1752	33.3333	11500	0.6426	0.4086
0.169	34.7826	12000	0.6633	0.4025
0.1612	36.2319	12500	0.6466	0.3968
0.1553	37.6812	13000	0.6573	0.3941
0.15	39.1304	13500	0.6989	0.3956
0.1442	40.5797	14000	0.7209	0.4062
0.1409	42.0290	14500	0.6950	0.3961
0.1356	43.4783	15000	0.6816	0.3863
0.134	44.9275	15500	0.6896	0.3867
0.1288	46.3768	16000	0.7073	0.3844
0.1263	47.8261	16500	0.7207	0.3836
0.1218	49.2754	17000	0.7430	0.3812
0.1217	50.7246	17500	0.7588	0.3831
0.1183	52.1739	18000	0.7478	0.3813
0.113	53.6232	18500	0.7269	0.3779
0.1109	55.0725	19000	0.7117	0.3735
0.1102	56.5217	19500	0.7532	0.3689
0.1084	57.9710	20000	0.7608	0.3704
0.1042	59.4203	20500	0.7571	0.3677
0.1048	60.8696	21000	0.7745	0.3683
0.1005	62.3188	21500	0.7845	0.3712
0.1006	63.7681	22000	0.7633	0.3664
0.0976	65.2174	22500	0.7721	0.3639
0.096	66.6667	23000	0.7659	0.3643
0.0938	68.1159	23500	0.7658	0.3620
0.0933	69.5652	24000	0.7692	0.3579
0.092	71.0145	24500	0.7785	0.3625
0.089	72.4638	25000	0.7845	0.3615
0.088	73.9130	25500	0.7973	0.3586
0.0862	75.3623	26000	0.7806	0.3576
0.0851	76.8116	26500	0.7947	0.3583
0.0846	78.2609	27000	0.7802	0.3526
0.0809	79.7101	27500	0.8093	0.3532
0.0813	81.1594	28000	0.8237	0.3572
0.0785	82.6087	28500	0.8130	0.3533
0.0799	84.0580	29000	0.7958	0.3511
0.0784	85.5072	29500	0.8108	0.3507
0.0767	86.9565	30000	0.8208	0.3511
0.0742	88.4058	30500	0.8270	0.3501
0.0746	89.8551	31000	0.8121	0.3459
0.073	91.3043	31500	0.8151	0.3485
0.0725	92.7536	32000	0.8265	0.3477
0.0717	94.2029	32500	0.8173	0.3446
0.0709	95.6522	33000	0.8135	0.3434
0.0704	97.1014	33500	0.8179	0.3431
0.0699	98.5507	34000	0.8134	0.3427
0.0691	100.0	34500	0.8155	0.3428

Framework versions

• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0