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: 0.5218
• Wer: 0.3434

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
3.5634	1.0	500	2.0727	1.0096
0.9357	2.01	1000	0.6623	0.5634
0.4536	3.01	1500	1.4421	0.4829
0.3044	4.02	2000	0.4361	0.4363
0.2369	5.02	2500	0.5098	0.4495
0.1994	6.02	3000	0.4741	0.3711
0.1699	7.03	3500	0.4652	0.3898
0.1499	8.03	4000	0.4151	0.3949
0.1308	9.04	4500	0.4685	0.3838
0.1234	10.04	5000	0.5076	0.3794
0.1055	11.04	5500	0.4492	0.3790
0.0953	12.05	6000	0.4726	0.3679
0.0863	13.05	6500	0.4797	0.3717
0.0816	14.06	7000	0.4725	0.3655
0.0842	15.06	7500	0.5181	0.3405
0.0661	16.06	8000	0.5315	0.3510
0.0593	17.07	8500	0.5024	0.3668
0.0624	18.07	9000	0.5374	0.3663
0.0535	19.08	9500	0.4861	0.3517
0.0524	20.08	10000	0.4812	0.3574
0.0461	21.08	10500	0.4976	0.3431
0.0363	22.09	11000	0.5062	0.3476
0.0351	23.09	11500	0.5094	0.3479
0.0327	24.1	12000	0.5291	0.3455
0.0319	25.1	12500	0.5209	0.3460
0.0268	26.1	13000	0.5173	0.3481
0.0263	27.11	13500	0.5362	0.3486
0.0234	28.11	14000	0.5333	0.3444
0.0237	29.12	14500	0.5218	0.3434

Framework versions

• Transformers 4.17.0
• Pytorch 1.12.1+cu113
• Datasets 1.18.3
• Tokenizers 0.13.0