ReVoiceAI-HuBERT-Thai-IPA

This model is a fine-tuned version of facebook/hubert-base-ls960 on the common_voice_17_0 (mapped to phonemic IPA) dataset. It achieves the following results on the evaluation set:

• Loss: -1.0097
• Wer: 0.2411
• Cer: 0.0791

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: 0.0005
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
-0.6917	1.0	936	-0.7912	0.4519	0.1585
-0.5931	2.0	1872	-0.8088	0.4398	0.1575
-0.5726	3.0	2808	-0.8250	0.4368	0.1507
-0.5981	4.0	3744	-0.8131	0.4270	0.1466
-0.5875	5.0	4680	-0.8349	0.4235	0.1435
-0.6103	6.0	5616	-0.8398	0.4353	0.1510
-0.628	7.0	6552	-0.8283	0.4069	0.1419
-0.6334	8.0	7488	-0.8420	0.3986	0.1353
-0.6452	9.0	8424	-0.8667	0.3889	0.1317
-0.6724	10.0	9360	-0.8883	0.3879	0.1341
-0.6735	11.0	10296	-0.8766	0.3996	0.1353
-0.6745	12.0	11232	-0.8851	0.3597	0.1231
-0.7151	13.0	12168	-0.9114	0.3450	0.1179
-0.7099	14.0	13104	-0.9199	0.3253	0.1107
-0.736	15.0	14040	-0.9214	0.3226	0.1083
-0.7644	16.0	14976	-0.9648	0.3106	0.1042
-0.7684	17.0	15912	-0.9583	0.3159	0.1060
-0.7815	18.0	16848	-0.9573	0.2921	0.0983
-0.8026	19.0	17784	-0.9406	0.3260	0.1073
-0.8239	20.0	18720	-0.9867	0.2845	0.0957
-0.8248	21.0	19656	-0.9921	0.2723	0.0903
-0.8463	22.0	20592	-1.0006	0.2632	0.0882
-0.8622	23.0	21528	-1.0012	0.2564	0.0849
-0.8485	24.0	22464	-1.0012	0.2558	0.0837
-0.856	25.0	23400	-1.0023	0.2570	0.0839
-0.8768	26.0	24336	-0.9986	0.2489	0.0820
-0.8987	27.0	25272	-1.0149	0.2416	0.0793
-0.8845	28.0	26208	-1.0041	0.2453	0.0802
-0.8812	29.0	27144	-1.0108	0.2409	0.0791
-0.8906	30.0	28080	-1.0097	0.2411	0.0791

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1
• Datasets 3.6.0
• Tokenizers 0.21.0