unispeech-sat-base-timit-ft

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

• Loss: 0.6712
• Wer: 0.4101

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
3.2582	0.69	100	3.1651	1.0
2.9542	1.38	200	2.9567	1.0
2.9656	2.07	300	2.9195	1.0
2.8946	2.76	400	2.8641	1.0
1.9305	3.45	500	1.7680	1.0029
1.0134	4.14	600	1.0184	0.6942
0.8355	4.83	700	0.7769	0.6080
0.8724	5.52	800	0.7182	0.6035
0.5619	6.21	900	0.6823	0.5406
0.4247	6.9	1000	0.6279	0.5237
0.4257	7.59	1100	0.6056	0.5000
0.5007	8.28	1200	0.5870	0.4918
0.3854	8.97	1300	0.6200	0.4804
0.264	9.66	1400	0.6030	0.4600
0.1989	10.34	1500	0.6049	0.4588
0.3196	11.03	1600	0.5946	0.4599
0.2622	11.72	1700	0.6282	0.4422
0.1697	12.41	1800	0.6559	0.4413
0.1464	13.1	1900	0.6349	0.4328
0.2277	13.79	2000	0.6133	0.4284
0.221	14.48	2100	0.6617	0.4219
0.1391	15.17	2200	0.6705	0.4235
0.112	15.86	2300	0.6207	0.4218
0.1717	16.55	2400	0.6749	0.4184
0.2081	17.24	2500	0.6756	0.4169
0.1244	17.93	2600	0.6750	0.4181
0.0978	18.62	2700	0.6500	0.4115
0.128	19.31	2800	0.6750	0.4106
0.1791	20.0	2900	0.6712	0.4101

Framework versions

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