wangpuupup/whisper-small-nl-dy-noaugmentation

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: 0.4891
• Wer: 22.8706

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
0.1493	2.02	500	0.3960	28.9779
0.0383	5.02	1000	0.4041	26.5132
0.0264	8.01	1500	0.4274	25.5890
0.0155	11.01	2000	0.4437	24.7735
0.0041	14.01	2500	0.4454	25.0453
0.0044	17.01	3000	0.4444	23.9761
0.0044	20.01	3500	0.4394	23.4868
0.0022	23.01	4000	0.4415	22.8525
0.0034	26.01	4500	0.4602	23.6499
0.0027	29.01	5000	0.4577	23.3780
0.0072	32.01	5500	0.4573	23.3962
0.0002	35.01	6000	0.4673	23.1062
0.0001	38.01	6500	0.4723	22.9975
0.0001	41.01	7000	0.4770	23.0881
0.0	44.01	7500	0.4807	23.0518
0.0	47.01	8000	0.4835	22.9612
0.0	50.01	8500	0.4857	22.9250
0.0	53.0	9000	0.4874	22.9069
0.0	56.0	9500	0.4887	22.9069
0.0	59.0	10000	0.4891	22.8706

Framework versions

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