jethrowang/webcam_loss_CH_0.5_emb-whisper-tiny

Whisper Tiny Hakka Simulated 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: 0.1428
• Cer: 9.1488

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 976
• training_steps: 9760
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.2406	0.9980	488	0.3252	23.5719
0.1309	1.9959	976	0.2390	24.1302
0.0836	2.9939	1464	0.2087	16.4008
0.0431	3.9918	1952	0.1886	18.9368
0.027	4.9898	2440	0.1932	12.6823
0.0173	5.9877	2928	0.1937	11.5218
0.0098	6.9857	3416	0.1835	9.9163
0.0094	7.9836	3904	0.1782	13.6325
0.0057	8.9816	4392	0.1806	14.1642
0.0051	9.9796	4880	0.1627	10.9358
0.0023	10.9775	5368	0.1680	9.8724
0.0025	11.9755	5856	0.1682	14.1179
0.0007	12.9734	6344	0.1537	9.5638
0.0007	13.9714	6832	0.1586	9.7637
0.0003	14.9693	7320	0.1432	8.6807
0.0001	15.9673	7808	0.1463	8.6865
0.0001	16.9652	8296	0.1445	8.1883
0.0001	17.9632	8784	0.1443	8.6044
0.0001	18.9611	9272	0.1427	8.9731
0.0001	19.9591	9760	0.1428	9.1488

Framework versions

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