librispeech-100h-supervised-meta

This model is a fine-tuned version of Kuray107/librispeech-5h-supervised on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0965
• Wer: 0.0330

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: 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: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1131	1.12	1000	0.0755	0.0487
0.0725	2.24	2000	0.0637	0.0404
0.0539	3.36	3000	0.0661	0.0389
0.0441	4.48	4000	0.0637	0.0371
0.0379	5.61	5000	0.0675	0.0356
0.0341	6.73	6000	0.0735	0.0360
0.0295	7.85	7000	0.0737	0.0362
0.0265	8.97	8000	0.0741	0.0350
0.0244	10.09	9000	0.0779	0.0337
0.0217	11.21	10000	0.0835	0.0343
0.0203	12.33	11000	0.0785	0.0339
0.0188	13.45	12000	0.0827	0.0344
0.0179	14.57	13000	0.0875	0.0332
0.0169	15.7	14000	0.0860	0.0330
0.0158	16.82	15000	0.0954	0.0330
0.0147	17.94	16000	0.0934	0.0329
0.0148	19.06	17000	0.0965	0.0330

Framework versions

• Transformers 4.16.2
• Pytorch 1.10.2
• Datasets 1.18.2
• Tokenizers 0.10.3