Sageleaf Beats Lab
MycoVox
The Mycelial Voice Protocol
An AI-powered, encrypted, self-sovereign music system. Vocal models trained locally, stored encrypted, with cryptographic consent built into every step.
No voice shall be harvested, only hosted. No model shall persist without consent. The right to sing includes the right to silence.
The Mission
To cultivate a fungal network of sovereign voice synthesis — decentralized, encrypted, ephemeral, and rooted in unforgeable biological consent. MycoVox ensures that vocal identity remains under the exclusive authority of its originator, resistant to extraction, surveillance, and corporate enclosure.
Core Principles
- • Data Dignity. Vocal identity is biometric data. It belongs to the person it came from — always. No third-party storage, no cloud residue.
- • Cryptographic Consent. Every voice model requires a GPG-signed consent manifest before training begins. Consent can be revoked. Models can be shredded on demand.
- • Ephemeral by Design. Models have retention policies baked in. After the project window closes, the model is cryptographically erased — overwritten seven times. A certificate of destruction is issued.
- • Air-Gapped Training. No cloud GPUs. No remote VRAM. Training runs on your hardware, in your room, with your rules.
The Stack
Voice Synthesis
DiffSinger — diffusion-based singing voice synthesis, trained locally
Voice Conversion
RVC — real-time retrieval-based voice conversion for inference
Cryptography
GnuPG for consent signing, LUKS2 full disk encryption for model storage
Hardware Security
YubiKey HSM for system encryption and model signing/attestation
"Like mycelium, the system operates as a rhizomatic network: horizontal, resilient, and capable of data exchange without centralized storage. The fruiting body — the generated audio — emerges from encrypted substrate, leaving no permanent spores in third-party soil."