wav2vec2-base-timit-fine-tuned

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

• Loss: 0.3457
• Wer: 0.2151

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 1
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 20.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.1621	0.69	100	3.1102	1.0
2.9592	1.38	200	2.9603	1.0
2.9116	2.07	300	2.8921	1.0
2.1332	2.76	400	1.9718	0.9958
0.8477	3.45	500	0.7813	0.5237
0.4251	4.14	600	0.5166	0.3982
0.3743	4.83	700	0.4400	0.3578
0.4194	5.52	800	0.4077	0.3370
0.23	6.21	900	0.4018	0.3142
0.1554	6.9	1000	0.3623	0.2995
0.1511	7.59	1100	0.3433	0.2697
0.1983	8.28	1200	0.3539	0.2715
0.1443	8.97	1300	0.3622	0.2551
0.0971	9.66	1400	0.3580	0.2519
0.0764	10.34	1500	0.3529	0.2437
0.1203	11.03	1600	0.3455	0.2431
0.0881	11.72	1700	0.3648	0.2415
0.0521	12.41	1800	0.3564	0.2320
0.0434	13.1	1900	0.3485	0.2270
0.0864	13.79	2000	0.3517	0.2228
0.0651	14.48	2100	0.3506	0.2285
0.0423	15.17	2200	0.3428	0.2247
0.0302	15.86	2300	0.3372	0.2198
0.0548	16.55	2400	0.3496	0.2196
0.0674	17.24	2500	0.3407	0.2166
0.0291	17.93	2600	0.3512	0.2171
0.0298	18.62	2700	0.3363	0.2158
0.0419	19.31	2800	0.3493	0.2145
0.046	20.0	2900	0.3457	0.2151

Framework versions

• Transformers 4.12.0.dev0
• Pytorch 1.8.1
• Datasets 1.14.1.dev0
• Tokenizers 0.10.3