Mirror VocalRender (base, pro) from pymaster/VocalRender@c2ba9d06 — bf16 model + fp32 audiovae safetensors, sha256 in NOTICE
aaa799a verified | license: apache-2.0 | |
| base_model: openbmb/VoxCPM2 | |
| tags: [singing-voice-synthesis, score-conditioned, music, maestraea] | |
| # VocalRender — Mæstræa mirror | |
| Mirror of [pymaster/VocalRender](https://huggingface.co/pymaster/VocalRender) | |
| (Apache-2.0) for the Mæstræa DAW's download-on-demand system. | |
| - `base/` — the VocalRender checkpoint (synthetic pretraining + CrawlSinger-OS finetune) | |
| - `pro/` — VocalRender-Pro (trained longer on the larger real-singing CrawlSinger corpus) | |
| Per variant: `model.safetensors` (bf16 cast of the upstream fp32 release), | |
| `audiovae.safetensors` (fp32, converted from `audiovae.pth`), tokenizer and | |
| config files. See NOTICE for source + output sha256 digests. | |