shimonbh/stt-small

Whisper Small He - Sanchit Gandhi

This model is a fine-tuned version of openai/whisper-small on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.9486
• Wer: 71.8172

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: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1591	5.1948	400	0.6390	74.0479
0.0164	10.3896	800	0.7373	74.3199
0.0078	15.5844	1200	0.8321	73.3950
0.0028	20.7792	1600	0.8744	74.2111
0.0012	25.9740	2000	0.9024	71.8172
0.0003	31.1688	2400	0.9137	72.0348
0.0002	36.3636	2800	0.9279	71.8172
0.0002	41.5584	3200	0.9391	71.8172
0.0002	46.7532	3600	0.9457	71.8172
0.0002	51.9481	4000	0.9486	71.8172

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.2.0
• Tokenizers 0.19.1