wav2vec2-large-xls-r-300m-j-phoneme-colab-new

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_10_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5498
• Wer: 0.3257

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: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• 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: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.0	397	0.7976	0.7045
No log	2.0	794	0.5777	0.5723
1.7064	3.0	1191	0.4775	0.4706
1.7064	4.0	1588	0.4755	0.4580
1.7064	5.0	1985	0.4678	0.4250
0.3823	6.0	2382	0.4742	0.4196
0.3823	7.0	2779	0.4419	0.3817
0.2485	8.0	3176	0.4402	0.3711
0.2485	9.0	3573	0.4942	0.3703
0.2485	10.0	3970	0.4877	0.3613
0.1735	11.0	4367	0.5073	0.3453
0.1735	12.0	4764	0.5127	0.3354
0.1238	13.0	5161	0.5545	0.3392
0.1238	14.0	5558	0.5419	0.3290
0.1238	15.0	5955	0.5498	0.3257

Framework versions

• Transformers 4.21.3
• Pytorch 1.10.0+cu113
• Datasets 2.4.0
• Tokenizers 0.12.1