jethrowang/webcam_loss_CH_1.0_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: 0.1845
• Cer: 9.7476

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
0.2444	0.9980	488	0.3550	25.2919
0.1293	1.9959	976	0.3005	32.6282
0.0835	2.9939	1464	0.2549	15.8437
0.0429	3.9918	1952	0.2720	23.6551
0.0279	4.9898	2440	0.2567	17.1440
0.0177	5.9877	2928	0.2180	13.6903
0.0103	6.9857	3416	0.2410	14.9398
0.007	7.9836	3904	0.2333	12.8846
0.0052	8.9816	4392	0.2008	12.4708
0.004	9.9796	4880	0.2393	12.1957
0.0023	10.9775	5368	0.2057	12.9829
0.0021	11.9755	5856	0.2098	12.4096
0.0014	12.9734	6344	0.1898	14.2208
0.001	13.9714	6832	0.2006	10.6954
0.0012	14.9693	7320	0.1941	13.2117
0.0005	15.9673	7808	0.1864	11.2548
0.0005	16.9652	8296	0.1887	10.7012
0.0001	17.9632	8784	0.1830	10.1197
0.0001	18.9611	9272	0.1863	9.8146
0.0001	19.9591	9760	0.1845	9.7476

Framework versions

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