I actually considered adding those types of voice lines, but since the voice was generated using GPT-SoVITS, it came with some limitations. The training samples were already quite high-pitched, and while it worked great for regular dialogue, it didn’t handle short, emotional sounds like screams or hit reactions very well. I also tried running those lines through So-VITS-SVC for voice conversion, but the results weren’t satisfying either—especially for sharp, abrupt expressions. In the end, I decided to leave them out for now to maintain the overall audio quality and consistency.
4 comments
I also tried running those lines through So-VITS-SVC for voice conversion, but the results weren’t satisfying either—especially for sharp, abrupt expressions. In the end, I decided to leave them out for now to maintain the overall audio quality and consistency.