Continue training after expanding dataset?

[skol101]

Say I've expanded my speaker's dataset with another hour of speech. The training on a pre-trained model has counted 20k steps so far.

Does it make sense continue training with a newly expanded dataset or I shall restart training?

[erogol]

To me makes sense to continue the training. Even maybe you can train a couple of epoch with only the new dataset to see how it affects the model.