Voice log

Conversation journal that uses Soniox realtime STT in a background daemon that:

• Captures microphone audio continuously.
• Keeps a text-only log file, with live conversation logs bucketed by minute.
• Keeps only the latest 60 minutes (for now).

Commands

Run from this skill directory:

npm install
node scripts/voice_journal_ctl.js start
node scripts/voice_journal_ctl.js end
node scripts/voice_journal_ctl.js status
node scripts/voice_journal_ctl.js last 10

OpenClaw trigger handling

When user says:

• start voice journal: run node scripts/voice_journal_ctl.js start.
• start voice log: run node scripts/voice_journal_ctl.js start.
• start voice log ["en","de"]: run node scripts/voice_journal_ctl.js start '["en","de"]'.
• end voice journal: run node scripts/voice_journal_ctl.js end.
• summarize what we talked about for last 10 minutes: run node scripts/voice_journal_ctl.js last 10, then summarize the returned text.

• Reply with only the requested outcome in one short sentence.
• Do not paste raw command output or transcript snippets unless the user explicitly asks for raw transcript/log text.
• If no text exists in range, report that explicitly.
• Never fabricate transcript text.

Required env

Set:

• SONIOX_API_KEY (required)
• Get/create key: https://soniox.com/speech-to-text

• None. Runtime settings are intentionally hard-coded except language hints passed in the start command.

Fixed defaults

• Data directory: ./.data under this skill.
• Soniox websocket endpoint: SDK default (SONIOX_API_WS_URL).
• Soniox model: stt-rt-v4.
• last output cap: 1800 chars by default, or override per command with --max-chars.
• Daemon environment: only SONIOX_API_KEY (and optional language hints) is forwarded; unrelated host env secrets are not inherited.

Audio capture defaults

Auto-selects available command by platform. Recommended:

• Linux: arecord -q -f S16_LE -r 16000 -c 1 -t raw
• macOS: sox -q -d -t raw -b 16 -e signed-integer -r 16000 -c 1 -