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: 0.5035
• Wer: 0.3346

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
1.1411	1.0	500	0.6675	0.6001
0.5668	2.01	1000	0.4699	0.4973
0.3773	3.01	1500	0.4475	0.4403
0.2696	4.02	2000	0.4162	0.4166
0.2165	5.02	2500	0.3809	0.4011
0.1849	6.02	3000	0.4120	0.3849
0.1542	7.03	3500	0.4436	0.3770
0.1385	8.03	4000	0.3977	0.3732
0.1224	9.04	4500	0.4530	0.3672
0.1233	10.04	5000	0.3949	0.3596
0.1078	11.04	5500	0.4616	0.3539
0.097	12.05	6000	0.4354	0.3697
0.0821	13.05	6500	0.4370	0.3643
0.0724	14.06	7000	0.4729	0.3587
0.0678	15.06	7500	0.5822	0.3742
0.0632	16.06	8000	0.4460	0.3513
0.0627	17.07	8500	0.5531	0.3537
0.0574	18.07	9000	0.5262	0.3575
0.0515	19.08	9500	0.4794	0.3488
0.0475	20.08	10000	0.4941	0.3458
0.0463	21.08	10500	0.4741	0.3377
0.0392	22.09	11000	0.5390	0.3381
0.0401	23.09	11500	0.4984	0.3413
0.0371	24.1	12000	0.5112	0.3460
0.0305	25.1	12500	0.5255	0.3418
0.0278	26.1	13000	0.5045	0.3389
0.0265	27.11	13500	0.4990	0.3371
0.0248	28.11	14000	0.5242	0.3362
0.0249	29.12	14500	0.5035	0.3346

Framework versions

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