wav2vec2-xls-r-300m_phoneme-timit_english_timit-4k_001

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

• Loss: 0.3668
• Per: 0.1149

Among the four models (wav2vec2-base, wav2vec2-large, wav2vec2-xls-r-300m, and 1b), the xls-r-300m achieved the lowest loss with one of the lowest Phone Error Rate. I advise you to use this model for TIMIT-format English phoneme recognition.

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 5000
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Per
6.9932	3.46	1000	3.8704	0.9878
3.2566	6.92	2000	1.0333	0.2395
0.8914	10.38	3000	0.4078	0.1547
0.5663	13.84	4000	0.3696	0.1694
0.4799	17.3	5000	0.3688	0.1584
0.4216	20.76	6000	0.3668	0.1524
0.367	24.22	7000	0.3715	0.1529
0.3187	27.68	8000	0.3840	0.1667
0.289	31.14	9000	0.3911	0.1608
0.262	34.6	10000	0.3987	0.1593

Framework versions

• Transformers 4.28.1
• Pytorch 2.0.1
• Datasets 2.18.0
• Tokenizers 0.13.3