CoolFace
Apppublic

bossbeats/ibis-founder-voice

sourceHugging Facemitupdated 13d agoView on Hugging Face
0likes
App README

IBIS Founder Voice — Chatterbox Nano

This is the FTN-owned primary founder-voice synthesis service. It uses the MIT-licensed Resemble AI Chatterbox Nano model and zero-shot voice cloning from the founder-authorized reference recording.

Privacy and trust contract

The raw reference recording is never stored in this repository or baked into the container image. Mount it at runtime as a private file and set IBIS_FOUNDER_VOICE_REFERENCE_PATH. At startup/use, the service refuses synthesis unless the recording's SHA-256 exactly matches the approved founder sample:

a1c58062344bb586dad665b9db6b81bba56af85fbffd214a6c17df3f4d270e9b

The service also requires a private bearer token via IBIS_FOUNDER_VOICE_SERVICE_TOKEN. The public browser must never call this service directly; ibis-founder-voice is the policy/rate-limit gateway.

Required environment

  • —IBIS_FOUNDER_VOICE_SERVICE_TOKEN — long random server-to-server token
  • —IBIS_FOUNDER_VOICE_REFERENCE_PATH — mounted private reference file path; defaults to /run/secrets/ibis-founder-voice/reference.ogg
  • —IBIS_FOUNDER_VOICE_REFERENCE_B64 — alternative private secret containing the approved six-second derived reference clip; preferred on Hugging Face Spaces
  • —PORT — default 8080
  • —optional TORCH_NUM_THREADS

Routes

  • —GET /health — authenticated, no generation; proves provider/model/reference identity and readiness.
  • —POST /speak with { "text": "..." } — authenticated; returns base64 WAV plus founder-voice provenance.

Deployment

The image is deliberately portable across CPU container hosts with enough memory for the model. Chatterbox Nano is selected with ChatterboxTurboTTS.from_pretrained(device="cpu", nano=True). The private reference is verified and the model is loaded during application startup, so /health cannot report ready before the runtime is genuinely usable. Model weights download at startup unless the host supplies a persistent Hugging Face cache.

For a Hugging Face ZeroGPU Gradio Space, add IBIS_FOUNDER_VOICE_SERVICE_TOKEN and IBIS_FOUNDER_VOICE_REFERENCE_B64 as Secrets, never Variables or repository files. space_app.py exposes hidden authenticated Gradio API endpoints and requests GPU capacity only for synthesis. The public Space contains no founder audio or credential.

Generate the compact secret from the authorized full recording with:

bash
bash derive-reference.sh /private/path/founder-reference.ogg /private/path/founder-reference-6s.ogg
base64 -w 0 /private/path/founder-reference-6s.ogg

The derivation script refuses a different source recording and produces a deterministic artifact matching the SHA-256 pinned in app.py.

The edge gateway should be configured with:

  • —IBIS_FOUNDER_VOICE_OPEN_URL=https://<private-or-protected-service>
  • —IBIS_FOUNDER_VOICE_OPEN_TOKEN=<same server-to-server token>

A generic TTS voice is never an acceptable substitute for IBIS_FOUNDER_VOICE.