By name: family-office-voice-stack id: SK-YSJ-0201 domain: yousirjuan status: Active description: > Sovereign voice stack — M5 Holler/Parakeet edge, DGX Whisper/Fish fallback, tower-api routing tiers, no cloud TTS/STT.
Family Office Voice Stack
When to use
Voice, TTS, STT, Parakeet pad, Jarvis lane, or "which engine for talk-and-reply?"
Architecture
| Tier | Engine | Host | Role |
|---|---|---|---|
| premium | Holler (Qwen3-TTS) | M5 Max | Daily Jarvis voice |
| premium | F5-TTS / NeMo Riva | DGX | Clone / prime quality |
| fallback | Kokoro / speaches | DGX | Emergency only |
| STT | faster-whisper | M5 | Edge route |
| STT | nephew-whisper | DGX | Brain route |
Jarvis = voice only. Visual = Obsidian. Never cite retired Kokoro as primary.
Operator commands (Nephew repo)
make ensure-m5-voice
make m5-voice-loop-probe
make cassette-line CHECK=voice
curl -s http://10.1.0.5:8088/api/v1/voice/tts # route probe via CT
Canonical docs
- YSJ
docs/setup/11-voice-parakeet-premium-stack.md - YSJ
docs/setup/28-voice-containers-whisper-fish-speech.md - Nephew
data/voice-config.json·docs/pockit/Parakeet-Voice-Cassette-Vanilla.md
Sovereign rule
No cloud speech APIs. NEPHEW_EGRESS_DENY=1 on voice-related deploy overlays.