seiching/whisper-large-seiching

Whisper large zh - seiching

This model is a fine-tuned version of openai/whisper-large on the Common Voice 13 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2457
• Wer Ortho: 40.3316
• Wer: 39.9281

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: 50
• training_steps: 4000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.0361	0.69	500	0.1989	38.3627	37.9517
0.0105	1.38	1000	0.2217	39.0259	38.9100
0.0208	2.06	1500	0.2299	39.6891	39.3292
0.0091	2.75	2000	0.2264	39.8964	39.4091
0.0153	3.44	2500	0.2363	39.8135	39.3891
0.0191	4.13	3000	0.2415	40.1865	40.0080
0.0061	4.81	3500	0.2542	41.1813	39.9281
0.0107	5.5	4000	0.2457	40.3316	39.9281

Framework versions

• Transformers 4.30.2
• Pytorch 1.13.1+cu117
• Datasets 2.13.2
• Tokenizers 0.13.3