jethrowang/whisper-tiny_tat-esc_exp_cc_0.5_embed

Whisper Tiny Taiwanese (exp_cc_0.5_embed)

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

• Loss: 2.3468
• Cer: 40.8566

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: 0.0005
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Use 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: 1362
• training_steps: 13620
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.4029	0.9985	681	1.0622	40.3475
0.3838	1.9971	1362	1.2521	45.3507
0.2719	2.9956	2043	1.2355	44.4585
0.1927	3.9941	2724	1.2278	44.0645
0.1391	4.9927	3405	1.2620	44.1542
0.0996	5.9912	4086	1.3637	43.2260
0.0691	6.9897	4767	1.4641	44.2366
0.053	7.9883	5448	1.4679	43.7573
0.0401	8.9868	6129	1.5647	42.3763
0.0331	9.9853	6810	1.6093	42.9464
0.0243	10.9839	7491	1.6949	43.3876
0.0186	11.9824	8172	1.7528	42.7885
0.0134	12.9809	8853	1.8851	43.8099
0.0093	13.9795	9534	1.9080	42.4202
0.0069	14.9780	10215	1.9732	42.1247
0.0039	15.9765	10896	2.1342	41.9910
0.0018	16.9751	11577	2.1268	42.0237
0.0005	17.9736	12258	2.2460	40.5185
0.0002	18.9721	12939	2.3131	40.9274
0.0001	19.9707	13620	2.3468	40.8566

Framework versions

• Transformers 4.49.0
• Pytorch 2.0.0.post304
• Datasets 3.3.2
• Tokenizers 0.21.0