Switch between MWW/OWW models from the UI (#348)

This PR adds the ability to select microWakeWord/openWakeWord models using the wake word dropdown selection via the HA device page at any time without the need for restarting the app/container to change the wake word path.

linux_voice_assistant/__main__.py

@@ -18,7 +18,7 @@ from getmac import get_mac_address  # type: ignore
 from pymicro_wakeword import MicroWakeWord, MicroWakeWordFeatures
 from pyopen_wakeword import OpenWakeWord, OpenWakeWordFeatures
 
-from .models import Preferences, ServerState
+from .models import Preferences, ServerState, WakeWordType
 from .mpv_player import MpvMediaPlayer
 from .peripheral_api import LVAEvent, PeripheralAPIServer
 from .satellite import VoiceSatelliteProtocol
@@ -322,6 +322,25 @@ async def main() -> None:
 
     # Load available wake words
     wake_word_dirs = [Path(ww_dir) for ww_dir in args.wake_word_dir]
+
+    # If the operator explicitly pointed --wake-word-dir (or the WAKE_WORD_DIR
+    # env var) at the openWakeWord subdirectory, prefer resolving --wake-model
+    # to an openWakeWord model of the same name instead of a same-named
+    # microWakeWord one. Checked before the automatic dirs below are appended,
+    # since those always include the openWakeWord path and would otherwise
+    # make every configuration look like an openWakeWord preference.
+    preferred_wake_word_type = WakeWordType.OPEN_WAKE_WORD if any("openwakeword" in str(ww_dir).lower() for ww_dir in wake_word_dirs) else None
+
+    # openWakeWord models ship in their own subdirectory under the default
+    # wakewords dir. find_available_wake_words() only globs the top level of
+    # each directory it's given, so this must be added explicitly or the OWW
+    # models never get discovered (and never show up in the HA dropdown).
+    # Appended after the user-specified dirs so OWW entries are inserted
+    # (and therefore displayed) after the microWakeWord ones.
+    oww_dir = _WAKEWORDS_DIR / "openWakeWord"
+    if oww_dir not in wake_word_dirs:
+        wake_word_dirs.append(oww_dir)
+
     wake_word_dirs.append(args.download_dir / "external_wake_words")
     available_wake_words = find_available_wake_words(wake_word_dirs, args.stop_model)
 
@@ -359,7 +378,12 @@ async def main() -> None:
         preferences.mic_noise_suppression = args.mic_noise_suppression
 
     # Load wake/stop models
-    wake_models, active_wake_words, fallback_used = load_wake_models(available_wake_words, [word for word in preferences.active_wake_words if word is not None], args.wake_model)
+    wake_models, active_wake_words, fallback_used = load_wake_models(
+        available_wake_words,
+        [word for word in preferences.active_wake_words if word is not None],
+        args.wake_model,
+        preferred_type=preferred_wake_word_type,
+    )
 
     # TODO: allow openWakeWord for "stop"
     stop_model = load_stop_model(wake_word_dirs, args.stop_model)

linux_voice_assistant/wake_word.py

@@ -76,8 +76,43 @@ def find_available_wake_words(wake_word_dirs: List[Path], stop_model_id: str) ->
     return available_wake_words
 
 
+def _find_matching_wake_word_id(
+    available_wake_words: Dict[str, AvailableWakeWord],
+    requested_id: str,
+    wake_word_type: Optional[WakeWordType] = None,
+) -> Optional[str]:
+    """
+    Finds a wake word id matching requested_id, optionally restricted to a single type.
+
+    Tries, in order:
+        1. An exact id match.
+        2. An id that starts with "{requested_id}_", which covers openWakeWord's
+           versioned filenames (e.g. "hey_jarvis" -> "hey_jarvis_v0.1").
+
+    Args:
+        available_wake_words: Dictionary with all available wake words
+        requested_id: ID (or base name) to look for
+        wake_word_type: If given, only consider wake words of this type
+
+    Returns:
+        The matching id, or None if no match was found
+    """
+    prefix = f"{requested_id}_"
+    for candidate_id, wake_word in available_wake_words.items():
+        if wake_word_type is not None and wake_word.type != wake_word_type:
+            continue
+
+        if candidate_id == requested_id or candidate_id.startswith(prefix):
+            return candidate_id
+
+    return None
+
+
 def load_wake_models(
-    available_wake_words: Dict[str, AvailableWakeWord], active_wake_word_ids: Optional[List[str]], default_wake_word_id: str
+    available_wake_words: Dict[str, AvailableWakeWord],
+    active_wake_word_ids: Optional[List[str]],
+    default_wake_word_id: str,
+    preferred_type: Optional[WakeWordType] = None,
 ) -> tuple[Dict[str, Union[MicroWakeWord, OpenWakeWord]], Set[str], bool]:
     """
     Loads the specified wake word models.
@@ -88,6 +123,12 @@ def load_wake_models(
         available_wake_words: Dictionary with all available wake words
         active_wake_word_ids: List of IDs of wake words to load (may be None)
         default_wake_word_id: ID of the default model which is loaded if no others are specified
+        preferred_type: If given, prefer resolving default_wake_word_id (and the
+            "okay_nabu" fallback) to a wake word of this type before falling back
+            to a type-agnostic match. Lets the operator's --wake-word-dir choice
+            (e.g. pointing at the openWakeWord subdirectory) decide which model
+            variant gets activated when the requested id exists for more than
+            one wake word engine.
 
     Returns:
         Tuple with (Dictionary of loaded models, Set of active wake word IDs)
@@ -120,14 +161,40 @@ def load_wake_models(
         # No models loaded, fall back to default model
         _LOGGER.debug("No wake models loaded, falling back to default model")
         wake_word_id = default_wake_word_id
+        wake_word = None
+
+        # If the operator's --wake-word-dir choice implies a preferred engine
+        # (e.g. pointing at the openWakeWord subdirectory), try to resolve the
+        # requested id to a model of that type first, even if a same-named
+        # model of a different type also exists.
+        if preferred_type is not None:
+            matched_id = _find_matching_wake_word_id(available_wake_words, default_wake_word_id, preferred_type)
+            if matched_id is not None:
+                wake_word_id = matched_id
+                wake_word = available_wake_words[wake_word_id]
+                _LOGGER.debug("Resolved default wake word '%s' to '%s' (preferred type %s)", default_wake_word_id, wake_word_id, preferred_type)
+
+        if wake_word is None:
+            # Fall back to a type-agnostic match (covers exact ids and
+            # versioned openWakeWord filenames like "hey_jarvis_v0.1")
+            matched_id = _find_matching_wake_word_id(available_wake_words, default_wake_word_id)
+            if matched_id is not None:
+                wake_word_id = matched_id
+                wake_word = available_wake_words[wake_word_id]
 
-        # Check if default wake word exists
-        wake_word = available_wake_words.get(wake_word_id)
         if wake_word is None:
             _LOGGER.error("❌ Default wake word '%s' not found!", wake_word_id)
 
-            # Try fallback to 'okay_nabu'
+            # Try fallback to 'okay_nabu', respecting the same type preference
             wake_word_id = "okay_nabu"
+            matched_id = None
+            if preferred_type is not None:
+                matched_id = _find_matching_wake_word_id(available_wake_words, wake_word_id, preferred_type)
+            if matched_id is None:
+                matched_id = _find_matching_wake_word_id(available_wake_words, wake_word_id)
+            if matched_id is not None:
+                wake_word_id = matched_id
+
             wake_word = available_wake_words.get(wake_word_id)
 
             if wake_word is None:

wakewords/openWakeWord/alexa_v0.1.json

@@ -1,6 +1,6 @@
 {
   "type": "openWakeWord",
-  "wake_word": "Alexa",
+  "wake_word": "Alexa (OWW)",
   "model": "alexa_v0.1.tflite",
   "openWakeWord": {
     "probability_cutoff": 0.7

wakewords/openWakeWord/hey_jarvis_v0.1.json

@@ -1,6 +1,6 @@
 {
   "type": "openWakeWord",
-  "wake_word": "Hey Jarvis",
+  "wake_word": "Hey Jarvis (OWW)",
   "model": "hey_jarvis_v0.1.tflite",
   "openWakeWord": {
     "probability_cutoff": 0.7

wakewords/openWakeWord/hey_mycroft_v0.1.json

@@ -1,6 +1,6 @@
 {
   "type": "openWakeWord",
-  "wake_word": "Hey Mycroft",
+  "wake_word": "Hey Mycroft (OWW)",
   "model": "hey_mycroft_v0.1.tflite",
   "openWakeWord": {
     "probability_cutoff": 0.7

wakewords/openWakeWord/hey_rhasspy_v0.1.json

@@ -1,6 +1,6 @@
 {
   "type": "openWakeWord",
-  "wake_word": "Hey Rhasspy",
+  "wake_word": "Hey Rhasspy (OWW)",
   "model": "hey_rhasspy_v0.1.tflite",
   "openWakeWord": {
     "probability_cutoff": 0.7

wakewords/openWakeWord/ok_nabu_v0.1.json

@@ -1,6 +1,6 @@
 {
   "type": "openWakeWord",
-  "wake_word": "Ok Nabu",
+  "wake_word": "Ok Nabu (OWW)",
   "model": "ok_nabu_v0.1.tflite",
   "openWakeWord": {
     "probability_cutoff": 0.7