The 100 epochs is more expresive, but less fidelity.
The 300 epochs is expresive and has good fidelity.
The 1025 epochs model is almost none expresive like the OG voice.