whisper-tiny-us-ZA

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

• Loss: 0.6915
• Wer Ortho: 0.2821
• Wer: 0.2784

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: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 5
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.3413	3.125	100	0.4281	0.2727	0.2474
0.0659	6.25	200	0.4672	0.2754	0.2526
0.0076	9.375	300	0.5252	0.3035	0.2899
0.0019	12.5	400	0.5568	0.2874	0.2758
0.0009	15.625	500	0.5804	0.2901	0.2771
0.0006	18.75	600	0.5947	0.2861	0.2732
0.0005	21.875	700	0.6062	0.2848	0.2745
0.0004	25.0	800	0.6170	0.2834	0.2745
0.0003	28.125	900	0.6261	0.2834	0.2745
0.0003	31.25	1000	0.6346	0.2781	0.2719
0.0002	34.375	1100	0.6423	0.2794	0.2732
0.0002	37.5	1200	0.6497	0.2794	0.2732
0.0002	40.625	1300	0.6563	0.2794	0.2732
0.0002	43.75	1400	0.6627	0.2794	0.2732
0.0001	46.875	1500	0.6680	0.2941	0.2874
0.0001	50.0	1600	0.6736	0.2874	0.2809
0.0001	53.125	1700	0.6781	0.2874	0.2809
0.0001	56.25	1800	0.6833	0.2874	0.2809
0.0001	59.375	1900	0.6876	0.2834	0.2796
0.0001	62.5	2000	0.6915	0.2821	0.2784

Framework versions

• Transformers 4.41.0.dev0
• Pytorch 2.1.2
• Datasets 2.18.0
• Tokenizers 0.19.1