english-filipino-wav2vec2-l-xls-r-test-06

This model is a fine-tuned version of jonatasgrosman/wav2vec2-large-xlsr-53-english on the filipino_voice dataset. It achieves the following results on the evaluation set:

• Loss: 1.5806
• Wer: 0.6568

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: 0.002
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.0031	2.09	400	1.2366	0.8780
0.9084	4.19	800	1.0653	0.8081
0.6484	6.28	1200	1.1648	0.8258
0.5335	8.38	1600	1.0903	0.7542
0.4359	10.47	2000	0.9466	0.7058
0.3629	12.57	2400	0.9266	0.7048
0.3057	14.66	2800	1.0879	0.7018
0.2477	16.75	3200	1.1113	0.7022
0.208	18.85	3600	1.1345	0.6742
0.1781	20.94	4000	1.3117	0.6974
0.1465	23.04	4400	1.3248	0.6916
0.1288	25.13	4800	1.4306	0.6523
0.1108	27.23	5200	1.5155	0.6685
0.099	29.32	5600	1.5806	0.6568

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu113
• Datasets 1.18.3
• Tokenizers 0.10.3