BALAKA/wav2vec2-large-xlsr-53-th-swear-words

wav2vec2-large-xlsr-53-th-main

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

• Loss: 1.1340
• Wer: 0.4686

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 30
• num_epochs: 40
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.2524	3.23	100	3.3222	1.0
3.2913	6.45	200	3.1818	1.0
2.222	9.68	300	1.2497	0.5335
1.1558	12.9	400	1.0792	0.5214
0.934	16.13	500	1.0663	0.4986
0.8023	19.35	600	1.0331	0.4893
0.7041	22.58	700	1.0801	0.4800
0.6576	25.81	800	1.1123	0.4886
0.6061	29.03	900	1.0748	0.4829
0.5649	32.26	1000	1.1187	0.4679
0.5717	35.48	1100	1.1267	0.4715
0.5267	38.71	1200	1.1340	0.4686

Framework versions

• Transformers 4.28.0
• Pytorch 2.0.1+cu118
• Datasets 1.16.1
• Tokenizers 0.13.3