wav2vec2-base-timit-demo-google-colab

This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5816
• Wer: 0.3533

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: 4
• 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: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.243	0.5	500	1.0798	0.7752
0.834	1.01	1000	0.6206	0.5955
0.5503	1.51	1500	0.5387	0.5155
0.4548	2.01	2000	0.4660	0.4763
0.3412	2.51	2500	0.8381	0.4836
0.3128	3.02	3000	0.4818	0.4519
0.2547	3.52	3500	0.4415	0.4230
0.2529	4.02	4000	0.4624	0.4219
0.2103	4.52	4500	0.4714	0.4096
0.2102	5.03	5000	0.4968	0.4087
0.1838	5.53	5500	0.4643	0.4131
0.1721	6.03	6000	0.4676	0.3979
0.1548	6.53	6500	0.4765	0.4085
0.1595	7.04	7000	0.4797	0.3941
0.1399	7.54	7500	0.4753	0.3902
0.1368	8.04	8000	0.4697	0.3945
0.1276	8.54	8500	0.5438	0.3869
0.1255	9.05	9000	0.5660	0.3841
0.1077	9.55	9500	0.4964	0.3947
0.1197	10.05	10000	0.5349	0.3849
0.1014	10.55	10500	0.5558	0.3883
0.0949	11.06	11000	0.5673	0.3785
0.0882	11.56	11500	0.5589	0.3955
0.0906	12.06	12000	0.5752	0.4120
0.1064	12.56	12500	0.5080	0.3727
0.0854	13.07	13000	0.5398	0.3798
0.0754	13.57	13500	0.5237	0.3816
0.0791	14.07	14000	0.4967	0.3725
0.0731	14.57	14500	0.5287	0.3744
0.0719	15.08	15000	0.5633	0.3596
0.062	15.58	15500	0.5399	0.3752
0.0681	16.08	16000	0.5151	0.3759
0.0559	16.58	16500	0.5564	0.3709
0.0533	17.09	17000	0.5933	0.3743
0.0563	17.59	17500	0.5381	0.3670
0.0527	18.09	18000	0.5685	0.3731
0.0492	18.59	18500	0.5728	0.3725
0.0509	19.1	19000	0.6074	0.3807
0.0436	19.6	19500	0.5762	0.3628
0.0434	20.1	20000	0.6721	0.3729
0.0416	20.6	20500	0.5842	0.3700
0.0431	21.11	21000	0.5374	0.3607
0.037	21.61	21500	0.5556	0.3667
0.036	22.11	22000	0.5608	0.3592
0.04	22.61	22500	0.5272	0.3637
0.047	23.12	23000	0.5234	0.3625
0.0506	23.62	23500	0.5427	0.3629
0.0418	24.12	24000	0.5590	0.3626
0.037	24.62	24500	0.5615	0.3555
0.0429	25.13	25000	0.5806	0.3616
0.045	25.63	25500	0.5777	0.3639
0.0283	26.13	26000	0.5987	0.3617
0.0253	26.63	26500	0.5671	0.3551
0.032	27.14	27000	0.5464	0.3582
0.0321	27.64	27500	0.5634	0.3573
0.0274	28.14	28000	0.5513	0.3575
0.0245	28.64	28500	0.5745	0.3537
0.0251	29.15	29000	0.5759	0.3547
0.0222	29.65	29500	0.5816	0.3533

Framework versions

• Transformers 4.17.0
• Pytorch 1.11.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1