feat(terra): luna v3 voice, channel marker and rail panel

- TTS defaults to eleven_v3_conversational over the Text to Dialogue
  socket: performs audio tags and gets her German right. Flash still
  works via the TTS socket.
- Every spoken message is prefixed with a [voice] marker so luna's
  profile on mars can tell the channel apart.
- LunaIndicator/LunaCall replaced by a LunaPanel in the bar rail, with
  captions; headless test updated to match.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-09-25 00:52:03 +02:00
co-authored by Claude Opus 5.5
parent 386a46c4fe
commit e4ee637ccd
13 changed files with 675 additions and 467 deletions
+28 -6
View File
@@ -1,11 +1,12 @@
{ config, pkgs, lib, ... }:
# Voice loop for luna: hold a key, talk, hear her answer.
# Voice loop for luna: SUPER+V opens the mic, talk, hear her answer.
#
# mic -> whisper-cpp (local, vulkan) -> Hermes' OpenAI-compatible API on mars
# -> ElevenLabs streaming TTS -> speaker. State, amplitude and the viseme
# timeline are published on a unix socket for the quickshell avatar to read
# (nothing consumes it yet — the widget is the next phase).
# -> ElevenLabs streaming TTS -> speaker. State, amplitude, captions and a
# viseme timeline are published on a unix socket; the rail's LunaPanel reads
# them (dotfiles/quickshell/HyprChrome/Widgets/Bar/Luna), and the viseme
# stream is there for the avatar that replaces its meter later.
#
# A USER service, not a system one: it needs darman's pipewire session and his
# XDG_RUNTIME_DIR for the socket. That is also why the two secrets it reads
@@ -57,8 +58,13 @@ in
ttsModel = lib.mkOption {
type = lib.types.str;
default = "eleven_flash_v2_5";
description = "~75ms model, the one ElevenLabs recommends for live conversation.";
default = "eleven_v3_conversational";
description = ''
eleven_v3* streams over the Text to Dialogue socket and performs audio
tags ([laughs], [sighs]); it also gets her German right, which Flash
reads with an English accent. Anything else uses the TTS socket, e.g.
"eleven_flash_v2_5" — half the price per character, faster, no tags.
'';
};
hermesKeyFile = lib.mkOption {
@@ -99,6 +105,21 @@ in
description = "Silence that ends an utterance.";
};
voiceMarker = lib.mkOption {
type = lib.types.str;
default = "[voice]";
description = ''
Prefixed to every spoken message so luna can tell this channel from
Telegram or the dashboard, and adjust — no kaomoji, no markdown, short
answers. Put the rule in HER profile on mars rather than only in
systemPrompt below: the marker travels with the message, so it still
identifies the channel in whatever history Hermes keeps, and one place
governs her voice manners across every client that sends it.
Empty string disables the marker.
'';
};
systemPrompt = lib.mkOption {
type = lib.types.nullOr lib.types.str;
default = null;
@@ -131,6 +152,7 @@ in
LUNA_VAD_MULT = toString cfg.vadMult;
LUNA_VAD_FLOOR = toString cfg.vadFloor;
LUNA_SILENCE_MS = toString cfg.silenceMs;
LUNA_VOICE_MARKER = cfg.voiceMarker;
} // lib.optionalAttrs (cfg.systemPrompt != null) {
LUNA_SYSTEM_PROMPT = cfg.systemPrompt;
} // lib.optionalAttrs (cfg.inputDevice != null) {