wav2vec2-common_voice-tr-demo

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: 0.3815
• Wer: 0.3493

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

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.92	100	3.5559	1.0
No log	1.83	200	3.0161	0.9999
No log	2.75	300	0.8587	0.7443
No log	3.67	400	0.5855	0.6121
3.1095	4.59	500	0.4841	0.5204
3.1095	5.5	600	0.4533	0.4923
3.1095	6.42	700	0.4157	0.4342
3.1095	7.34	800	0.4304	0.4334
3.1095	8.26	900	0.4097	0.4068
0.2249	9.17	1000	0.4049	0.3881
0.2249	10.09	1100	0.3993	0.3809
0.2249	11.01	1200	0.3855	0.3782
0.2249	11.93	1300	0.3923	0.3713
0.2249	12.84	1400	0.3833	0.3591
0.1029	13.76	1500	0.3811	0.3570
0.1029	14.68	1600	0.3834	0.3499

Framework versions

• Transformers 4.13.0.dev0
• Pytorch 1.12.0a0+2c916ef
• Datasets 2.2.2
• Tokenizers 0.10.3