Zzzkay1/whisper-small-nan

Whisper Small Taiwanese (Min Nan)

This model is a fine-tuned version of openai/whisper-small on the Taiwanese Songs (Min Nan) dataset. It achieves the following results on the evaluation set:

• Loss: 2.6304
• Wer: 25.6671

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: 2
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.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: 200
• training_steps: 10000
• mixed_precision_training: Native AMP
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.5742	2.3284	100	3.5014	66.2008
1.9713	4.6569	200	2.6350	45.9975
1.5884	6.9853	300	2.4657	36.0864
1.5118	9.3050	400	2.5327	34.5616
1.4857	11.6334	500	2.5065	33.7992
1.4703	13.9619	600	2.4858	30.7497
1.459	16.2815	700	2.4334	28.3355
1.4511	18.6100	800	2.4682	28.9708
1.4472	20.9384	900	2.4897	29.3520
1.4435	23.2581	1000	2.5269	42.8208
1.4377	25.5865	1100	2.5061	26.9377
1.4364	27.9150	1200	2.5459	26.9377
1.4328	30.2346	1300	2.5174	28.0813
1.4334	32.5630	1400	2.5495	26.1753
1.429	34.8915	1500	2.5535	26.1753
1.427	37.2111	1600	2.5565	26.6836
1.4254	39.5396	1700	2.5540	25.0318
1.4241	41.8680	1800	2.5546	26.9377
1.423	44.1877	1900	2.5845	26.3024
1.4224	46.5161	2000	2.5926	26.4295
1.4215	48.8446	2100	2.6072	26.1753
1.4212	51.1642	2200	2.6304	25.6671

Framework versions

• Transformers 4.56.2
• Pytorch 2.7.1+cu118
• Datasets 4.1.1
• Tokenizers 0.22.1