fluent-clean-wav2vec

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

• Loss: 0.0100
• Wer: 0.2638

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
4.7739	1.26	500	2.7988	1.0
1.4369	2.53	1000	0.2079	0.5323
0.2838	3.79	1500	0.0565	0.3471
0.1845	5.05	2000	0.0435	0.3209
0.1383	6.31	2500	0.0284	0.3011
0.1131	7.58	3000	0.4893	0.2964
0.1127	8.84	3500	0.0340	0.2702
0.0942	10.1	4000	0.0155	0.2732
0.0779	11.36	4500	0.0134	0.2667
0.0665	12.63	5000	0.0130	0.2732
0.0619	13.89	5500	0.0163	0.2667
0.0539	15.15	6000	0.0514	0.2650
0.0456	16.41	6500	0.0110	0.2662
0.0405	17.68	7000	0.0105	0.2667
0.0343	18.94	7500	0.0297	0.2667
0.0325	20.2	8000	0.0109	0.2656
0.0241	21.46	8500	0.0109	0.2662
0.0214	22.73	9000	0.0136	0.2644
0.0215	23.99	9500	0.0101	0.2638
0.0215	25.25	10000	0.0101	0.2667
0.0226	26.52	10500	0.0096	0.2638
0.012	27.78	11000	0.0091	0.2644
0.0111	29.04	11500	0.0100	0.2638

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2