Transfer learning with conditioning

[aleksas]

My scenario is following:

• Train model with 9 speakers
• Use trained model to fine-tune for other nine speakers with arbitrary number of utterances.

Can fine-tuning instruction as indicated here be used for scenario above? Or it doesn't make sense.

The reason I'm asking is that I'm trying this scenario, but losses converge much slower.

P.S. Decided to change speakers, due to poor initial choice - old speakers had less utterances and one speaker had noise level too high.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.