jethrowang/whisper-tiny_tat-esc_exp_nr_0.5_cc_0.5_embeds

Whisper Tiny Taiwanese (exp_nr_0.5_cc_0.5_embeds)

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: 2.2774
• Cer: 41.7092

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: 681
• training_steps: 6810
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.4532	0.9985	681	1.3229	45.0006
0.2812	1.9971	1362	1.3009	47.7935
0.1813	2.9956	2043	1.2902	45.8631
0.119	3.9941	2724	1.3410	45.0435
0.0751	4.9927	3405	1.4026	43.7097
0.0409	5.9912	4086	1.6134	44.5456
0.0231	6.9897	4767	1.7609	42.9457
0.0094	7.9883	5448	1.9361	42.7805
0.0026	8.9868	6129	2.1500	41.6526
0.0005	9.9853	6810	2.2774	41.7092

Framework versions

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