Nova_HD_by_Kady
Trained on: 8m33s of clean, high-quality mono audio (48kHz), with diverse pitch and tone expression.
- Framework: Applio (RVC)
- Architecture: RVC v2
- Feature extractor: contentvec
- Pitch extraction: rmvpe
- Vocoder: HiFi-GAN
- Steps: 42k
- Epochs: 1000
- Batch size: 6
- Pretrain: contentvec hifigan / 48k
- Format:
.pth (with index + config.json), ONNX version also included
Details
While many recommend training RVC models for 200–300 epochs, Nova_HD was pushed to 1000 epochs intentionally. The result is a dramatic improvement in clarity, pitch accuracy, and tonal depth — especially compared to the lighter 300-epoch pre-existing model.
Despite the extended training, overfitting was not an issue. In fact, Nova's unique radio static voice coloration was preserved and enhanced, delivering a distinct synthetic quality without compromising vocal clarity.
Perfect for character speech, synthetic voiceovers, and AI avatars requiring a crisp and stylized tone.
📇 Metadata
- Author: Kady
- Discord:
kady_x
- Huggingface:
Kady-x
- Model version:
Nova_HD
- License: CC-BY 4.0
- Attribution: Voice model created by Kady (
Nova_HD)
