dysarthria-base

This model is a fine-tuned version of openai/whisper-base on an dysarthria dataset. It achieves the following results on the evaluation set:

• Loss: 0.2523
• Wer: 109.4374
• Cer: 72.5101

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0039	13.5135	500	0.3016	86.2069	60.2961
0.0003	27.0270	1000	0.2616	86.5699	65.8816
0.0001	40.5405	1500	0.2577	79.8548	63.3580
0.0001	54.0541	2000	0.2565	119.6007	78.6003
0.0001	67.5676	2500	0.2549	122.6860	80.0471
0.0001	81.0811	3000	0.2539	122.6860	80.0808
0.0	94.5946	3500	0.2533	117.7858	77.2207
0.0	108.1081	4000	0.2527	112.7042	74.8654
0.0	121.6216	4500	0.2523	109.4374	72.5774
0.0	135.1351	5000	0.2523	109.4374	72.5101

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0