wangpuupup/whisper-small-nl-dy-augmentation

Whisper Small dysarthric Dutch

This model is a fine-tuned version of qmeeus/whisper-small-nl on the data/copas copas-full dataset. It achieves the following results on the evaluation set:

• Loss: 0.4702
• Wer: 22.1638

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: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1618	0.05	500	0.3787	28.9235
0.0583	1.05	1000	0.3732	25.7702
0.0382	2.05	1500	0.4001	25.4621
0.0316	3.05	2000	0.4081	24.7010
0.0169	4.05	2500	0.4325	24.1935
0.0153	5.05	3000	0.4325	33.4179
0.0074	6.05	3500	0.4367	23.9398
0.0096	7.05	4000	0.4390	23.3055
0.0054	8.05	4500	0.4441	23.7042
0.0032	9.04	5000	0.4493	23.2693
0.004	10.04	5500	0.4524	23.3418
0.0048	11.04	6000	0.4498	23.7224
0.001	12.04	6500	0.4577	22.8887
0.0002	13.04	7000	0.4577	22.0913
0.0001	14.04	7500	0.4616	22.1276
0.0001	15.04	8000	0.4639	22.2726
0.0001	16.04	8500	0.4662	22.1095
0.0001	17.04	9000	0.4684	22.1457
0.0001	18.04	9500	0.4697	22.1457
0.0001	19.04	10000	0.4702	22.1638

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.12.1+cu116
• Datasets 2.4.0
• Tokenizers 0.12.1