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: 0.5079
• Wer: 0.3365

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: 8
• 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
3.4933	1.0	500	1.7711	0.9978
0.8658	2.01	1000	0.6262	0.5295
0.4405	3.01	1500	0.4841	0.4845
0.3062	4.02	2000	0.4897	0.4215
0.233	5.02	2500	0.4326	0.4101
0.1896	6.02	3000	0.4924	0.4078
0.1589	7.03	3500	0.4430	0.3896
0.1391	8.03	4000	0.4334	0.3889
0.1216	9.04	4500	0.4691	0.3828
0.1063	10.04	5000	0.4726	0.3705
0.0992	11.04	5500	0.4333	0.3690
0.0872	12.05	6000	0.4986	0.3771
0.0829	13.05	6500	0.4903	0.3685
0.0713	14.06	7000	0.5293	0.3655
0.068	15.06	7500	0.5039	0.3612
0.0621	16.06	8000	0.5314	0.3665
0.0571	17.07	8500	0.5038	0.3572
0.0585	18.07	9000	0.4718	0.3550
0.0487	19.08	9500	0.5482	0.3626
0.0459	20.08	10000	0.5239	0.3545
0.0419	21.08	10500	0.5096	0.3473
0.0362	22.09	11000	0.5222	0.3500
0.0331	23.09	11500	0.5062	0.3489
0.0352	24.1	12000	0.4913	0.3459
0.0315	25.1	12500	0.4701	0.3412
0.028	26.1	13000	0.5178	0.3402
0.0255	27.11	13500	0.5168	0.3405
0.0228	28.11	14000	0.5154	0.3368
0.0232	29.12	14500	0.5079	0.3365

Framework versions

• Transformers 4.17.0
• Pytorch 1.12.1+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1