fluent-noisy-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: 0.0129
• Wer: 0.2656

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: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.5477	1.26	500	2.9258	1.0
1.6916	2.53	1000	0.4439	0.5218
0.4069	3.79	1500	0.0990	0.3524
0.2584	5.05	2000	0.0812	0.3256
0.1954	6.31	2500	0.0340	0.2825
0.1391	7.58	3000	0.0691	0.3046
0.1378	8.84	3500	0.0334	0.2848
0.1088	10.1	4000	0.0349	0.2871
0.0972	11.36	4500	0.0959	0.2761
0.0883	12.63	5000	0.0229	0.2726
0.0734	13.89	5500	0.0303	0.2772
0.0644	15.15	6000	0.0251	0.2755
0.0536	16.41	6500	0.0139	0.2714
0.0428	17.68	7000	0.0214	0.2685
0.0362	18.94	7500	0.0196	0.2667
0.0377	20.2	8000	0.0257	0.2691
0.0289	21.46	8500	0.0191	0.2673
0.0297	22.73	9000	0.0207	0.2667
0.029	23.99	9500	0.0129	0.2656

Framework versions

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