wav2vec2-base-librispeech-model

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the LIBRI10H - ENG dataset. It achieves the following results on the evaluation set:

• Loss: 0.8515
• Wer: 0.7226

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 200
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.7426	1.1565	200	2.8968	1.0
2.7493	2.3130	400	2.2712	0.9987
2.0118	3.4696	600	1.6905	0.9768
1.7815	4.6261	800	1.5406	0.9588
1.667	5.7826	1000	1.4410	0.9385
1.5898	6.9391	1200	1.3799	0.9282
1.5366	8.0928	1400	1.3415	0.9165
1.4917	9.2493	1600	1.3144	0.9205
1.455	10.4058	1800	1.2746	0.9068
1.4266	11.5623	2000	1.2521	0.9102
1.3925	12.7188	2200	1.2213	0.8971
1.3754	13.8754	2400	1.2028	0.8938
1.3452	15.0290	2600	1.1931	0.8826
1.3265	16.1855	2800	1.1682	0.8860
1.3106	17.3420	3000	1.1645	0.8752
1.2917	18.4986	3200	1.1686	0.8780
1.2745	19.6551	3400	1.1385	0.8670
1.2639	20.8116	3600	1.1301	0.8666
1.2432	21.9681	3800	1.1173	0.8670
1.2294	23.1217	4000	1.1098	0.8620
1.2203	24.2783	4200	1.1077	0.8711
1.2037	25.4348	4400	1.0964	0.8626
1.1965	26.5913	4600	1.0910	0.8581
1.181	27.7478	4800	1.0842	0.8533
1.1711	28.9043	5000	1.0692	0.8465
1.1573	30.0580	5200	1.0724	0.8464
1.1472	31.2145	5400	1.0529	0.8404
1.1375	32.3710	5600	1.0506	0.8403
1.1276	33.5275	5800	1.0432	0.8398
1.1149	34.6841	6000	1.0371	0.8330
1.1099	35.8406	6200	1.0372	0.8341
1.0959	36.9971	6400	1.0296	0.8370
1.0838	38.1507	6600	1.0136	0.8232
1.0761	39.3072	6800	1.0355	0.8288
1.069	40.4638	7000	1.0072	0.8211
1.0624	41.6203	7200	1.0019	0.8217
1.0502	42.7768	7400	1.0021	0.8329
1.0423	43.9333	7600	0.9960	0.8153
1.0334	45.0870	7800	0.9903	0.8134
1.0203	46.2435	8000	0.9787	0.8116
1.0212	47.4	8200	0.9690	0.8029
1.0062	48.5565	8400	0.9864	0.8030
1.0029	49.7130	8600	0.9658	0.8000
0.9922	50.8696	8800	0.9552	0.7964
0.9784	52.0232	9000	0.9563	0.7978
0.9761	53.1797	9200	0.9442	0.7898
0.9649	54.3362	9400	0.9495	0.7898
0.9567	55.4928	9600	0.9448	0.7927
0.9556	56.6493	9800	0.9303	0.7851
0.9454	57.8058	10000	0.9304	0.7784
0.9356	58.9623	10200	0.9202	0.7718
0.927	60.1159	10400	0.9264	0.7730
0.9172	61.2725	10600	0.9252	0.7736
0.9177	62.4290	10800	0.9087	0.7682
0.9107	63.5855	11000	0.9119	0.7663
0.9017	64.7420	11200	0.9014	0.7609
0.899	65.8986	11400	0.8962	0.7597
0.8854	67.0522	11600	0.8976	0.7533
0.8841	68.2087	11800	0.8952	0.7554
0.8792	69.3652	12000	0.8951	0.7535
0.8697	70.5217	12200	0.8913	0.7513
0.8677	71.6783	12400	0.8820	0.7496
0.862	72.8348	12600	0.8834	0.7447
0.8573	73.9913	12800	0.8824	0.7437
0.8527	75.1449	13000	0.8747	0.7388
0.8451	76.3014	13200	0.8806	0.7399
0.8435	77.4580	13400	0.8713	0.7401
0.8393	78.6145	13600	0.8734	0.7387
0.8353	79.7710	13800	0.8702	0.7367
0.834	80.9275	14000	0.8661	0.7335
0.8265	82.0812	14200	0.8642	0.7312
0.8183	83.2377	14400	0.8638	0.7334
0.8238	84.3942	14600	0.8643	0.7311
0.8176	85.5507	14800	0.8640	0.7309
0.8183	86.7072	15000	0.8603	0.7294
0.8121	87.8638	15200	0.8586	0.7270
0.8033	89.0174	15400	0.8585	0.7265
0.8116	90.1739	15600	0.8560	0.7254
0.8058	91.3304	15800	0.8553	0.7262
0.7992	92.4870	16000	0.8548	0.7263
0.7979	93.6435	16200	0.8528	0.7236
0.7979	94.8	16400	0.8529	0.7235
0.7978	95.9565	16600	0.8526	0.7242
0.7934	97.1101	16800	0.8519	0.7238
0.7915	98.2667	17000	0.8520	0.7233
0.7996	99.4232	17200	0.8515	0.7230

Framework versions

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