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

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: 0.5212
• Wer: 0.2998

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• 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
0.7364	3.0	2000	0.4703	0.4503
0.5673	6.01	4000	0.4585	0.3855
0.5048	9.01	6000	0.4567	0.3543
0.4567	12.01	8000	0.4433	0.3473
0.4194	15.02	10000	0.4491	0.3386
0.3905	18.02	12000	0.4829	0.3360
0.3644	21.02	14000	0.5032	0.3306
0.3441	24.02	16000	0.5242	0.3389
0.2589	27.03	18000	0.5212	0.2998

Framework versions

• Transformers 4.22.2
• Pytorch 1.10.0+cu113
• Datasets 2.5.1
• Tokenizers 0.12.1