Do you have any tips on creating a pre-trained model to improve the performance of Speaker’s Emotion Maintenance? (emotion transfer)

[junylee11]

Do you have any tips on creating a pre-trained model to improve the performance of Speaker’s Emotion Maintenance? (emotion transfer)
(How to synthesize voice using emotion in reference wav file)

Learning was conducted with 12 hours of data, 100+100 steps, and made plbert, and Speaker’s Emotion Maintenance was performed on the produced model, but it was impossible to produce a voice containing the emotion and cadence of 'reference wav'.

I'm not sure if this is a problem with the amount of data, a problem with plbert or a problem with the learning step. Anyone please answer.