jethrowang/webcam_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: 0.1323
• Cer: 7.3156

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
0.2495	0.9980	488	0.3384	23.3014
0.1352	1.9959	976	0.2472	16.6320
0.0854	2.9939	1464	0.2133	16.6389
0.0447	3.9918	1952	0.1956	22.4831
0.0282	4.9898	2440	0.1921	12.4096
0.0167	5.9877	2928	0.1670	10.9682
0.0115	6.9857	3416	0.1833	10.5590
0.0078	7.9836	3904	0.1591	9.6204
0.0057	8.9816	4392	0.1568	10.5324
0.0037	9.9796	4880	0.1684	10.1371
0.0036	10.9775	5368	0.1626	10.8352
0.0015	11.9755	5856	0.1451	10.4226
0.0016	12.9734	6344	0.1562	10.2099
0.0007	13.9714	6832	0.1575	8.7731
0.0004	14.9693	7320	0.1395	9.8597
0.0006	15.9673	7808	0.1421	8.3316
0.0003	16.9652	8296	0.1345	7.5433
0.0001	17.9632	8784	0.1322	7.8692
0.0001	18.9611	9272	0.1326	7.3630
0.0001	19.9591	9760	0.1323	7.3156

Framework versions

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