VoiceServer reads voices from settings.json only, so the DA speaks with two voices and every notify logs "fallback defaults"

[lgertel]

Version: LifeOS 7.40.4 (ce046f2)

What

LIFEOS_CONFIG.toml is where the DA's voice is declared ([da.voices.main], required by LifeosConfig.ts), and hooks/lib/identity.ts already reads identity through loadLifeosConfig(). The VoiceServer never does: it loads voices from ~/.claude/settings.json daidentity.voices, a section the setup flow no longer writes. Result on a configured install:

• hook-driven lines (the "DA says" notifications) use the configured voice;
• every skill's curl -X POST /notify (hundreds per day) resolves no voice, logs Voice settings: fallback defaults (no config found for main) and speaks with the hard-coded default voice;
• the user hears two different voices for one assistant.

Where

LIFEOS/PULSE/VoiceServer/voice.ts

• header comment lines 9–10: "look up in settings.json daidentity.voices … use settings.json daidentity.voices.main as default";
• loadVoiceConfigFromSettings (line 209): join(homedir(), ".claude", "settings.json") (line 210), settings.daidentity || {} (line 220), daidentity.voices || {} (line 221); no read of LIFEOS_CONFIG.toml;
• line 566: log("warn", \Voice settings: fallback defaults (no config found for ${voice})`)`.

LIFEOS/TOOLS/LifeosConfig.ts lines 166–168 define and require [da.voices.main].voice_id; hooks/lib/identity.ts line 37 loads through it.

Reproduce

1. Set [da.voices.main] voice_id = "<any ElevenLabs id>" in LIFEOS_CONFIG.toml; leave settings.json without a daidentity block (the current setup output).
2. Start Pulse; curl -s -X POST localhost:31337/notify -H 'Content-Type: application/json' -d '{"message":"test"}'.
3. Read the VoiceServer log: fallback defaults (no config found for main); the audio plays in the default voice, not the configured one. Trigger a hook notification: the configured voice plays.

Expected

One source of truth for voices: the VoiceServer resolves main and named voices from LIFEOS_CONFIG.toml through LifeosConfig.ts, the same way hooks/lib/identity.ts does, with settings.json at most a legacy fallback; no "fallback defaults" warning on a configured install.

Fix shape

voice.ts builds LoadedVoiceConfig from loadLifeosConfig().da.voices (voice id, name, model, stability per entry; main as the default), keeping the settings.json path only when the TOML has no voices. One function replaced; a PR can follow.

[danielmiessler]

Landed: the voice server reads voices from LIFEOS_CONFIG.toml like the hooks do, with settings.json as the legacy fallback, at both the loader and the live read. Next release. Thanks.