jethrowang/webcam_loss_CH_0.2_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: 0.1485
• Cer: 9.8655

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
0.2194	0.9980	488	0.3440	23.2448
0.1223	1.9959	976	0.2657	17.5914
0.0757	2.9939	1464	0.2101	14.1260
0.0421	3.9918	1952	0.2134	21.8647
0.026	4.9898	2440	0.1855	13.0384
0.0149	5.9877	2928	0.2038	17.5579
0.0089	6.9857	3416	0.1944	12.3148
0.007	7.9836	3904	0.1885	13.8983
0.0065	8.9816	4392	0.1778	12.3136
0.0041	9.9796	4880	0.1649	15.9096
0.0026	10.9775	5368	0.1737	10.8930
0.0021	11.9755	5856	0.1648	12.6361
0.0011	12.9734	6344	0.1535	9.9903
0.0009	13.9714	6832	0.1586	11.9807
0.0008	14.9693	7320	0.1550	9.0332
0.0005	15.9673	7808	0.1535	8.8702
0.0003	16.9652	8296	0.1527	9.6019
0.0002	17.9632	8784	0.1501	9.7741
0.0001	18.9611	9272	0.1487	9.4043
0.0001	19.9591	9760	0.1485	9.8655

Framework versions

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