wav2vec2-common_voice-tr-output

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

• Loss: 0.3776
• Wer: 0.3243

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: 20.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.92	100	3.6020	1.0
No log	1.83	200	2.9971	0.9999
No log	2.75	300	0.9174	0.7772
No log	3.67	400	0.5668	0.6356
3.1619	4.59	500	0.4949	0.5256
3.1619	5.5	600	0.4516	0.4744
3.1619	6.42	700	0.4291	0.4575
3.1619	7.34	800	0.4330	0.4273
3.1619	8.26	900	0.4016	0.4145
0.2261	9.17	1000	0.4214	0.4005
0.2261	10.09	1100	0.4093	0.3946
0.2261	11.01	1200	0.4051	0.3917
0.2261	11.93	1300	0.3908	0.3719
0.2261	12.84	1400	0.3850	0.3603
0.1119	13.76	1500	0.3967	0.3645
0.1119	14.68	1600	0.3821	0.3526
0.1119	15.6	1700	0.3919	0.3519
0.1119	16.51	1800	0.3763	0.3366
0.1119	17.43	1900	0.3682	0.3349
0.074	18.35	2000	0.3753	0.3323
0.074	19.27	2100	0.3753	0.3267

Framework versions

• Transformers 4.28.1
• Pytorch 1.12.1+cu102
• Datasets 2.12.0
• Tokenizers 0.13.3