wtimit-base-960h-normal30percent-all

This model is a fine-tuned version of facebook/wav2vec2-base-960h on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8097
• Wer: 0.3692

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: 8
• 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: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.32	1.3889	1000	0.3506	0.2804
0.2342	2.7778	2000	0.4413	0.2977
0.183	4.1667	3000	0.4847	0.3134
0.1389	5.5556	4000	0.5576	0.3291
0.1156	6.9444	5000	0.6021	0.3405
0.1003	8.3333	6000	0.6778	0.3632
0.0921	9.7222	7000	0.6309	0.3549
0.0771	11.1111	8000	0.7765	0.3823
0.0674	12.5	9000	0.7512	0.3722
0.0629	13.8889	10000	0.6964	0.3764
0.0575	15.2778	11000	0.8090	0.3812
0.0531	16.6667	12000	0.8377	0.3919
0.044	18.0556	13000	0.8246	0.3881
0.0427	19.4444	14000	0.8331	0.3826
0.0415	20.8333	15000	0.8166	0.3800
0.0356	22.2222	16000	0.8550	0.3916
0.0359	23.6111	17000	0.7968	0.3843
0.0311	25.0	18000	0.8020	0.3788
0.0251	26.3889	19000	0.8026	0.3684
0.0264	27.7778	20000	0.7937	0.3743
0.0248	29.1667	21000	0.8097	0.3692

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1