jethrowang/vanilla-whisper-tiny

Whisper Tiny Hakka Condenser

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

• Loss: 0.2504
• Cer: 15.3594

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: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 488
• training_steps: 4880
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.5588	0.9980	488	0.7965	38.4319
0.1839	1.9959	976	0.3925	20.8417
0.1089	2.9939	1464	0.3106	16.9418
0.0759	3.9918	1952	0.2813	16.3904
0.0525	4.9898	2440	0.2622	16.1962
0.039	5.9877	2928	0.2545	15.5559
0.0271	6.9857	3416	0.2556	15.3501
0.0231	7.9836	3904	0.2504	14.6439
0.0196	8.9816	4392	0.2502	15.0311
0.0149	9.9796	4880	0.2504	15.3594

Framework versions

• Transformers 4.42.3
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1