Sonorus OmniVoice AudioVAE
This repository contains the VAE-only GGUF used by Sonorus and the
voxcpm2-upscaler branch of omnivoice.cpp to convert native OmniVoice output
to 48 kHz audio.
File
| File | Size | SHA-256 |
|---|---|---|
voxcpm2-audiovae-f16.gguf | 187,868,032 bytes | a5fb091c0a95172bdee2ee7230335dac7d3dc318d77ca100f095d023cabd5d97 |
The file was converted from the official
OpenBMB/VoxCPM2
audiovae.pth checkpoint using
tools/convert-voxcpm2-audiovae.py from the
omnivoice.cpp fork.
The original model and weights are Apache-2.0. Conversion provenance is also
embedded in the GGUF metadata.
This is only the AudioVAE. The OmniVoice language model and tokenizer GGUFs remain hosted in Serveurperso/OmniVoice-GGUF.