jethrowang/webcam_topline-whisper-tiny

Whisper Tiny Hakka Webcam

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: 0.0706
• Cer: 3.8803

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: 1521
• training_steps: 15210
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.3507	0.9993	1521	0.3448	18.5531
0.1263	1.9987	3042	0.1418	7.4508
0.0781	2.9980	4563	0.1043	5.2685
0.0547	3.9974	6084	0.0880	5.2373
0.0369	4.9967	7605	0.0800	4.4617
0.0247	5.9961	9126	0.0752	4.1670
0.0196	6.9954	10647	0.0721	4.0884
0.014	7.9947	12168	0.0707	3.6676
0.0103	8.9941	13689	0.0706	3.6757
0.0086	9.9934	15210	0.0706	3.8803

Framework versions

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