speech_ocean_hubert_mdd

This model is a fine-tuned version of facebook/hubert-large-ll60k on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2027
• Wer: 0.0517
• Cer: 0.0499

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
42.7069	0.9873	39	36.7247	0.9992	0.9977
16.2787	2.0	79	7.8315	1.0	1.0
6.7896	2.9873	118	4.5645	1.0	1.0
4.0104	4.0	158	3.8654	1.0	1.0
3.8037	4.9873	197	3.8060	1.0	1.0
3.7898	6.0	237	3.7695	1.0	1.0
3.7777	6.9873	276	3.7717	1.0	1.0
3.7442	8.0	316	3.7320	1.0	1.0
3.7286	8.9873	355	3.6978	1.0	1.0
3.6272	10.0	395	3.5089	1.0	1.0
3.0921	10.9873	434	2.6068	0.9992	0.9997
2.2556	12.0	474	1.6832	0.5880	0.6815
1.7791	12.9873	513	1.2117	0.3861	0.4433
1.2731	14.0	553	0.7338	0.1793	0.1505
0.9596	14.9873	592	0.4892	0.1220	0.1005
0.7152	16.0	632	0.3525	0.0892	0.0752
0.521	16.9873	671	0.2843	0.0704	0.0623
0.4791	18.0	711	0.2351	0.0607	0.0568
0.3992	18.9873	750	0.2120	0.0547	0.0523
0.4245	19.7468	780	0.2027	0.0517	0.0499

Framework versions

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