speech_ocean_wav2vec_mdd

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3663
• Wer: 0.0863
• Cer: 0.0692

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
45.149	0.9873	39	45.0584	1.0258	0.7932
40.7325	2.0	79	32.0660	1.0	1.0
14.8164	2.9873	118	8.1694	1.0	1.0
5.6535	4.0	158	4.5922	1.0	1.0
3.9508	4.9873	197	3.8581	1.0	1.0
3.8065	6.0	237	3.7907	1.0	1.0
3.766	6.9873	276	3.7769	1.0	1.0
3.7552	8.0	316	3.7465	1.0	1.0
3.7489	8.9873	355	3.7611	1.0	1.0
3.7263	10.0	395	3.7234	1.0	1.0
3.7343	10.9873	434	3.6934	1.0	1.0
3.6327	12.0	474	3.4204	1.0	1.0
3.1861	12.9873	513	2.7907	0.9710	0.9864
2.2814	14.0	553	1.7142	0.5088	0.5401
1.6854	14.9873	592	1.0573	0.2488	0.1914
1.2968	16.0	632	0.7282	0.1786	0.1391
0.8626	16.9873	671	0.5435	0.1305	0.0999
0.7852	18.0	711	0.4440	0.1046	0.0831
0.6332	18.9873	750	0.3847	0.0936	0.0748
0.6518	19.7468	780	0.3663	0.0863	0.0692

Framework versions

• Transformers 4.40.0
• Pytorch 2.2.1+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1