wav2vec2_common_voice_accents_us

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

• Loss: 0.2722

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: 48
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 384
• total_eval_batch_size: 32
• 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
4.549	1.28	400	0.8521
0.4066	2.56	800	0.2407
0.2262	3.83	1200	0.2070
0.1828	5.11	1600	0.2134
0.1565	6.39	2000	0.2060
0.1448	7.67	2400	0.2100
0.1333	8.95	2800	0.2036
0.121	10.22	3200	0.2192
0.1146	11.5	3600	0.2154
0.1108	12.78	4000	0.2223
0.1017	14.06	4400	0.2331
0.094	15.34	4800	0.2257
0.0896	16.61	5200	0.2229
0.0825	17.89	5600	0.2229
0.0777	19.17	6000	0.2417
0.0719	20.45	6400	0.2433
0.0659	21.73	6800	0.2447
0.0651	23.0	7200	0.2446
0.0587	24.28	7600	0.2542
0.056	25.56	8000	0.2587
0.0521	26.84	8400	0.2640
0.0494	28.12	8800	0.2753
0.0465	29.39	9200	0.2722

Framework versions

• Transformers 4.17.0
• Pytorch 1.10.2+cu102
• Datasets 1.18.4
• Tokenizers 0.11.6