neddamj/whisper-tiny-patois

Whisper tiny - Patois

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

• Loss: 1.6927
• Wer: 0.8005

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: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• 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
1.6981	2.6490	400	1.7500	1.2964
1.0133	5.2980	800	1.4418	0.9557
0.7521	7.9470	1200	1.3821	0.8719
0.5125	10.5960	1600	1.4081	0.7962
0.3403	13.2450	2000	1.4613	0.8105
0.2737	15.8940	2400	1.5187	0.7902
0.1871	18.5430	2800	1.5868	0.7868
0.146	21.1921	3200	1.6405	0.7881
0.13	23.8411	3600	1.6760	0.8007
0.1154	26.4901	4000	1.6927	0.8005

Framework versions

• Transformers 4.52.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1