whisper-tiny-minds14-us-vickymm

This model is a fine-tuned version of openai/whisper-tiny on the minds14-us (whisper-tiny) dataset. It achieves the following results on the evaluation set:

• Loss: 2.4049
• Wer: 0.7485
• Wer Ortho: 0.7569

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 500
• training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Ortho
1.1149	1.79	100	0.5379	0.4097	0.4176
0.1705	3.57	200	0.7637	0.5762	0.5836
0.166	5.36	300	1.2479	0.5384	0.5416
0.2409	7.14	400	1.5261	0.6765	0.6619
0.2773	8.93	500	1.8106	0.7863	0.7816
0.2715	10.71	600	2.0421	0.7739	0.7841
0.2434	12.5	700	2.2664	0.7456	0.7514
0.1979	14.29	800	2.1956	0.6983	0.7039
0.1843	16.07	900	2.3711	0.8182	0.8229
0.1555	17.86	1000	2.4049	0.7485	0.7569

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.2
• Datasets 2.16.1
• Tokenizers 0.15.0