wav2vec2-base-common-voice-persian-colab

This model is a fine-tuned version of facebook/wav2vec2-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.1446
• Wer: 0.6911

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.26	300	3.0670	1.0
3.3475	2.52	600	2.5530	1.0
3.3475	3.78	900	1.4598	0.9555
2.0348	5.04	1200	1.2189	0.8797
1.0817	6.3	1500	1.1242	0.8268
1.0817	7.56	1800	1.0764	0.7957
0.7973	8.82	2100	1.1023	0.7863
0.7973	10.08	2400	1.0583	0.7785
0.6514	11.34	2700	1.0963	0.7512
0.5878	12.61	3000	1.1200	0.7494
0.5878	13.87	3300	1.0396	0.7402
0.484	15.13	3600	1.1407	0.7340
0.484	16.39	3900	1.1534	0.7584
0.4384	17.65	4200	1.0973	0.7236
0.3966	18.91	4500	1.0623	0.7358
0.3966	20.17	4800	1.1655	0.7112
0.3408	21.43	5100	1.1825	0.7084
0.3408	22.69	5400	1.1436	0.7029
0.3274	23.95	5700	1.1077	0.6988
0.2948	25.21	6000	1.1454	0.7066
0.2948	26.47	6300	1.1411	0.6956
0.2545	27.73	6600	1.0952	0.6918
0.2545	28.99	6900	1.1446	0.6911

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu113
• Datasets 1.18.3
• Tokenizers 0.10.3