---
license: mit
---

GPT-SoVITS v2 models trained on approx 140h of so-called "aegigoe" voice. May have ability to zero-shot. The "aegi" ability depends on reference audio.