wav2vec2-common_voice-ta

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the MOZILLA-FOUNDATION/COMMON_VOICE_6_1 - TA dataset. It achieves the following results on the evaluation set:

• Loss: 0.6563
• Wer: 0.7096

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_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: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.84	100	4.3941	1.0
No log	1.69	200	3.2005	1.0
No log	2.53	300	2.7844	1.0145
No log	3.38	400	0.8691	1.0003
4.317	4.22	500	0.6846	0.8394
4.317	5.06	600	0.6270	0.7790
4.317	5.91	700	0.5935	0.7802
4.317	6.75	800	0.5701	0.7812
4.317	7.59	900	0.5649	0.7891
0.3656	8.44	1000	0.6092	0.8178
0.3656	9.28	1100	0.6093	0.7721
0.3656	10.13	1200	0.6154	0.7287
0.3656	10.97	1300	0.6284	0.7408
0.3656	11.81	1400	0.6343	0.7143
0.1681	12.66	1500	0.6523	0.7363
0.1681	13.5	1600	0.6543	0.7139
0.1681	14.35	1700	0.6599	0.7094

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.17.0
• Tokenizers 0.15.1