wangpuupup/whisper_test

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: 0.4242
• Wer: 24.5560

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: 1e-05
• 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
0.3363	2.02	500	0.3762	29.7934
0.0945	5.02	1000	0.3418	27.6912
0.0332	8.01	1500	0.3353	26.1689
0.0147	11.01	2000	0.3476	26.1327
0.0071	14.01	2500	0.3623	25.9333
0.0034	17.01	3000	0.3789	25.2084
0.0024	20.01	3500	0.3827	24.8641
0.0026	23.01	4000	0.3877	25.3171
0.0021	26.01	4500	0.3933	25.4259
0.0014	29.01	5000	0.3941	25.0997
0.0008	32.01	5500	0.4014	25.0997
0.0004	35.01	6000	0.4035	24.8278
0.0003	38.01	6500	0.4080	24.9184
0.0003	41.01	7000	0.4120	24.8097
0.0002	44.01	7500	0.4151	24.6104
0.0002	47.01	8000	0.4176	24.3929
0.0002	50.01	8500	0.4200	24.5198
0.0001	53.0	9000	0.4230	24.5198
0.0001	56.0	9500	0.4252	24.4291
0.0001	59.0	10000	0.4242	24.5560

Framework versions

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