Voice

Dictation and read-aloud run on the owner server with faster-whisper and Piper — no paid speech API.

Mic → TranscribeAudio / /open-bot/voice/transcribe. TTS → Piper via synthesize/preview. Models live under /srv/openclaw/assistant-stack/voice with CPUQuota 150% / MemoryMax 2G.

Install: bash ops/voice/install.sh. Live voice-call websockets remain unwired (501). See repo docs/VOICE.md.