english-filipino-wav2vec2-l-xls-r-test-02

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: 0.4561
• Wer: 0.2632

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: 0.0003
• 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: 40
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.1707	2.09	400	0.8006	0.8224
0.4801	4.19	800	0.3363	0.4329
0.2541	6.28	1200	0.3365	0.3676
0.1851	8.38	1600	0.3485	0.3739
0.1408	10.47	2000	0.3628	0.3420
0.1098	12.57	2400	0.3979	0.3277
0.1019	14.66	2800	0.4031	0.2896
0.0887	16.75	3200	0.3977	0.3024
0.0798	18.85	3600	0.3959	0.3129
0.0671	20.94	4000	0.4489	0.3241
0.0633	23.04	4400	0.4455	0.3026
0.055	25.13	4800	0.4668	0.2910
0.0523	27.23	5200	0.4670	0.2960
0.0468	29.32	5600	0.4536	0.2781
0.0392	31.41	6000	0.4612	0.2860
0.0381	33.51	6400	0.4651	0.2841
0.034	35.6	6800	0.4723	0.2716
0.0315	37.7	7200	0.4546	0.2642
0.0294	39.79	7600	0.4561	0.2632

Framework versions

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