jethrowang/whisper-tiny_tat-esc_topline

Whisper Tiny Taiwanese (topline)

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: 1.0400
• Cer: 20.3105

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: 0.0001
• 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
0.6758	0.9985	681	0.6816	35.6624
0.5019	1.9971	1362	0.6077	24.6575
0.3484	2.9956	2043	0.5875	23.5932
0.2444	3.9941	2724	0.6029	22.1000
0.16	4.9927	3405	0.6502	22.3178
0.1084	5.9912	4086	0.7111	22.3447
0.0728	6.9897	4767	0.7801	22.1145
0.0493	7.9883	5448	0.8294	22.0905
0.0333	8.9868	6129	0.8626	22.4998
0.0248	9.9853	6810	0.8916	21.6134
0.018	10.9839	7491	0.9241	21.7539
0.0122	11.9824	8172	0.9620	21.7042
0.0086	12.9809	8853	0.9697	21.6206
0.0064	13.9795	9534	0.9937	21.1544
0.0037	14.9780	10215	1.0012	21.0531
0.0021	15.9765	10896	1.0125	20.6351
0.0015	16.9751	11577	1.0279	20.4550
0.0015	17.9736	12258	1.0328	20.2847
0.0018	18.9721	12939	1.0392	20.3533
0.0011	19.9707	13620	1.0400	20.3105

Framework versions

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