whisper-small-sv-extra-data

This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_13_0 and the google/fleurs datasets. It achieves the following results on the evaluation set:

• Loss: 0.3843
• Wer: 22.1585

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: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3189	0.53	500	0.3610	25.2029
0.1451	1.05	1000	0.3337	23.6495
0.1432	1.58	1500	0.3263	22.8908
0.0572	2.1	2000	0.3284	22.1622
0.0502	2.63	2500	0.3405	22.2000
0.0259	3.15	3000	0.3596	22.1924
0.0246	3.68	3500	0.3650	22.2208
0.0137	4.2	4000	0.3843	22.1585

Framework versions

• Transformers 4.36.0.dev0
• Pytorch 2.1.1+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0