ljcamargo/tachiwin_tutunaku

tachiwin_totonac

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

• Loss: 1.7535
• Wer: 0.6465

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: 90

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.1063	5.19	200	2.9834	1.0
2.9016	10.39	400	2.4405	0.9959
1.7606	15.58	600	1.1942	0.8532
1.0549	20.78	800	1.1132	0.7788
0.7553	25.97	1000	1.1224	0.6899
0.6639	31.51	1200	1.2641	0.7082
0.5344	36.7	1400	1.3247	0.6835
0.4527	41.9	1600	1.3915	0.7022
0.3839	47.09	1800	1.4051	0.6791
0.3065	52.29	2000	1.3899	0.6706
0.2714	57.48	2200	1.5455	0.6573
0.2437	62.68	2400	1.6798	0.6601
0.2103	67.87	2600	1.7406	0.6674
0.1899	73.06	2800	1.7625	0.6522
0.1841	78.26	3000	1.7443	0.6535
0.1544	83.45	3200	1.7405	0.6465
0.1461	88.65	3400	1.7535	0.6465

Framework versions

• Transformers 4.31.0
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3