jethrowang/android_emb-whisper-tiny

Whisper Tiny Taiwanese Simulated Android

This model is a fine-tuned version of openai/whisper-tiny on the TAT ASR Aligned dataset. It achieves the following results on the evaluation set:

• Loss: 0.7397
• Cer: 11.2806

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: 1362
• training_steps: 13620
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.3641	0.9985	681	0.4668	19.0185
0.2569	1.9971	1362	0.4366	14.5059
0.1682	2.9956	2043	0.4342	13.5919
0.1095	3.9941	2724	0.4588	13.0167
0.0693	4.9927	3405	0.4854	12.6401
0.0455	5.9912	4086	0.5303	13.1776
0.0323	6.9897	4767	0.5626	12.8424
0.0228	7.9883	5448	0.5940	12.4495
0.0168	8.9868	6129	0.6214	12.4219
0.0124	9.9853	6810	0.6661	13.1648
0.0091	10.9839	7491	0.6534	12.1909
0.0067	11.9824	8172	0.6671	12.1441
0.0036	12.9809	8853	0.6948	12.0141
0.0016	13.9795	9534	0.6962	11.7995
0.0011	14.9780	10215	0.7180	11.6767
0.0008	15.9765	10896	0.7170	11.5896
0.0005	16.9751	11577	0.7260	11.5133
0.0002	17.9736	12258	0.7299	11.3793
0.0002	18.9721	12939	0.7373	11.2399
0.0001	19.9707	13620	0.7397	11.2806

Framework versions

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