custom keyword - transfer learning

[Aishaj]

Hello,

I want to use this trained model as a base model; then, I will retrain the model with a custom dataset. The word that I want to train the model on is not included in the original commands dataset. Will this transfer learning process work ? The number of utterances for my keyword is not big.

Thank you

[Burton2000]

Yes this should be possible, you might need to modify the training code a little bit to load the pre-trained checkpoint. Also when loading weights you should try passing in skip_mismatch=True to avoid TF throwing an error when loading the model weights as you will have changed the model structure if you are training for a different number of keywords.

If doing transfer learning I don't think you would need too many samples in your dataset to get decent results.