jethrowang/webcam_loss_CH_1.0-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: 0.1884
• Cer: 9.2679

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
0.2208	0.9980	488	0.3739	25.8594
0.1188	1.9959	976	0.2960	24.9116
0.0782	2.9939	1464	0.2618	18.5519
0.041	3.9918	1952	0.2452	18.9357
0.0251	4.9898	2440	0.2292	17.5810
0.0169	5.9877	2928	0.2557	15.4137
0.011	6.9857	3416	0.2254	17.0585
0.0072	7.9836	3904	0.2343	12.5136
0.0051	8.9816	4392	0.2362	12.3864
0.0044	9.9796	4880	0.2261	12.0570
0.0024	10.9775	5368	0.2191	11.3219
0.0024	11.9755	5856	0.2158	12.1056
0.0012	12.9734	6344	0.2027	9.9140
0.0008	13.9714	6832	0.2002	9.7973
0.0003	14.9693	7320	0.2084	10.1822
0.001	15.9673	7808	0.1990	9.9082
0.0002	16.9652	8296	0.1946	9.4355
0.0001	17.9632	8784	0.1909	9.1361
0.0001	18.9611	9272	0.1901	9.0517
0.0001	19.9591	9760	0.1884	9.2679

Framework versions

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