is it possible to generate our own models?

[jlmxyz]

whisper.cpp is using pre-trained models, I wonder if it's possible to enhance the models by training them with more samples and different voices?