5.2.8. Chatterbox (TTS engine)

Chatterbox is a TTS engine selected with TTS_PROVIDER_SOURCE. See the TTS Provider README for setup and configuration.

1. Engine quick reference

• Image: travisvn/chatterbox-tts-api:gpu, pinned by digest (GPU only; voice-cloning model)
• License: AGPL-3.0 for the chatterbox-tts-api server; MIT for the Resemble AI model
• Activation: TTS_PROVIDER_SOURCE=chatterbox-container-gpu (or chatterbox-localhost for a host-installed instance)
• In-container port: 4123 (compose sets PORT=4123; the image's own default is 5123)
• Model cache: the chatterbox-cache volume at /cache (the image's MODEL_CACHE_DIR)
• Host port: ${CHATTERBOX_PORT} (computed from BASE_PORT by the bootstrapper)

2. Dependencies & Integrations

2.1. Current — Upstream (this service calls)

No upstream calls.

2.2. Current — Downstream (services that call this)

Service	Category
kong	infra
hermes	agents
n8n	agents
jupyterhub	apps
open-webui	apps

2.3. Architecture diagram

Open the full-size diagram for a full-screen view.

2.4. Future — Missing pair integrations

No high-confidence opportunities identified.

2.5. Future — Candidate new services

No high-confidence opportunities identified.

2.6. Future — Unused features in this service

No high-confidence opportunities identified.

3. Capabilities & limitations

Support tier: experimental — Capability contract declared; no cited cold-start, workflow, or upgrade qualification run yet (evidence at v0.1.0).

Capability	Status	Verification	Notes
GPU voice-cloning text-to-speech	supported	documented	The TTS selector starts the digest-pinned NVIDIA container and exposes Chatterbox synthesis and voice cloning through the selected provider endpoint.
Operator-run localhost Chatterbox	partial	documented	Atlas resolves a host Chatterbox endpoint selected by tts-provider, but installation, model downloads, process lifecycle, and hardware acceleration remain operator-owned.
Persistent registered voice library	not-supported	documented	The container persists only the Hugging Face weight cache; registered voice samples have no Atlas-managed volume or object-store workflow and may disappear on replacement.
Authenticated Chatterbox ingress	not-supported	documented	The host-published API and CORS-only tts.localhost Kong route have no Atlas authentication; use loopback or firewall controls, remove the publish, or add an authentication proxy.