unispeech-sat-base-plus-timit-ft

This model is a fine-tuned version of microsoft/unispeech-sat-base-plus on the TIMIT_ASR - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.6549
• Wer: 0.4051

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: 1
• 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: 20.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.3838	0.69	100	3.2528	1.0
2.9608	1.38	200	2.9682	1.0
2.9574	2.07	300	2.9346	1.0
2.8555	2.76	400	2.7612	1.0
1.7418	3.45	500	1.5732	0.9857
0.9606	4.14	600	1.0014	0.7052
0.8334	4.83	700	0.7691	0.6161
0.852	5.52	800	0.7169	0.5997
0.5707	6.21	900	0.6821	0.5527
0.4235	6.9	1000	0.6078	0.5140
0.4357	7.59	1100	0.5927	0.4982
0.5004	8.28	1200	0.5814	0.4826
0.3757	8.97	1300	0.5951	0.4643
0.2579	9.66	1400	0.5990	0.4581
0.2087	10.34	1500	0.5864	0.4488
0.3155	11.03	1600	0.5836	0.4464
0.2701	11.72	1700	0.6045	0.4348
0.172	12.41	1800	0.6494	0.4344
0.1529	13.1	1900	0.5915	0.4241
0.2411	13.79	2000	0.6156	0.4246
0.2348	14.48	2100	0.6363	0.4206
0.1429	15.17	2200	0.6394	0.4161
0.1151	15.86	2300	0.6186	0.4167
0.1723	16.55	2400	0.6498	0.4124
0.1997	17.24	2500	0.6541	0.4076
0.1297	17.93	2600	0.6546	0.4117
0.101	18.62	2700	0.6471	0.4075
0.1272	19.31	2800	0.6586	0.4065
0.1901	20.0	2900	0.6549	0.4051

Framework versions

• Transformers 4.12.0.dev0
• Pytorch 1.8.1
• Datasets 1.14.1.dev0
• Tokenizers 0.10.3