Fix: audio_output Hang + Stimme auf Seraphina

- speak() hing weil pygame.time.wait() die asyncio Event-Loop blockierte.
  Fix: time.sleep(0.05) statt pygame.time.wait() im Playback-Loop.
- asyncio.new_event_loop() statt asyncio.run() — thread-sicher, kein
  "event loop already running" in Worker-Threads.
- _speak_lock verhindert parallele TTS-Aufrufe.
- Stimme: de-DE-SeraphinaMultilingualNeural als Default (modern, weiblich,
  multilingual neural). Dropdown in Settings mit allen DE/EN Optionen.
- speak()-Signatur: voice + rate als Parameter (statt config-Import).

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

client/hermes-voice/audio_output.py

@@ -2,47 +2,59 @@ import asyncio
 import os
 import tempfile
 import threading
+import time
 import pygame
 import edge_tts
-from config import TTS_VOICE, TTS_RATE, TTS_PITCH
 
 pygame.mixer.init()
 _stop_event = threading.Event()
+_speak_lock = threading.Lock()
 
 
 def stop_speaking():
-    """Unterbricht laufende Sprachausgabe (z.B. wenn User spricht)."""
     _stop_event.set()
     pygame.mixer.music.stop()
 
 
-def speak(text: str):
+def speak(text: str, voice: str = "de-DE-SeraphinaMultilingualNeural", rate: str = "+0%"):
     """Text → Edge TTS → Lautsprecher. Blockiert bis fertig oder unterbrochen."""
+    with _speak_lock:
         _stop_event.clear()
 
-    async def _run():
+        tmp_path = None
+        try:
             with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as f:
                 tmp_path = f.name
 
-        communicate = edge_tts.Communicate(text, TTS_VOICE, rate=TTS_RATE, pitch=TTS_PITCH)
+            # Edge TTS in eigenem Event-Loop (Thread-sicher)
-        await communicate.save(tmp_path)
+            loop = asyncio.new_event_loop()
+            try:
+                communicate = edge_tts.Communicate(text, voice, rate=rate)
+                loop.run_until_complete(communicate.save(tmp_path))
+            finally:
+                loop.close()
 
             if _stop_event.is_set():
-            os.unlink(tmp_path)
                 return
 
             pygame.mixer.music.load(tmp_path)
             pygame.mixer.music.play()
 
+            # Polling ohne pygame.time.wait (blockiert Event-Loop nicht)
             while pygame.mixer.music.get_busy():
                 if _stop_event.is_set():
                     pygame.mixer.music.stop()
                     break
-            pygame.time.wait(50)
+                time.sleep(0.05)
 
+        except Exception:
+            pass
+        finally:
+            if tmp_path and os.path.exists(tmp_path):
+                try:
                     os.unlink(tmp_path)
-
+                except OSError:
-    asyncio.run(_run())
+                    pass
 
 
 def play_ding():
@@ -55,4 +67,4 @@ def play_ding():
     stereo = np.column_stack([wave, wave])
     sound = pygame.sndarray.make_sound(stereo)
     sound.play()
-    pygame.time.wait(int(duration * 1000) + 20)
+    time.sleep(duration + 0.02)

client/hermes-voice/config_manager.py

@@ -9,7 +9,7 @@ DEFAULTS = {
     "chat_model": "Hermes-4-14B",
     "vision_model": "Qwen3-VL-2B-Instruct",
     "hotkey": "ctrl_r",
-    "tts_voice": "de-DE-KillianNeural",
+    "tts_voice": "de-DE-SeraphinaMultilingualNeural",
     "tts_rate": "+0%",
     "whisper_model": "small",
     "language": "de",

client/hermes-voice/main.py

@@ -259,8 +259,13 @@ class SettingsWindow(ctk.CTkToplevel):
 
         label("TTS Stimme")
         self._voice = dropdown("tts_voice", [
-            "de-DE-KillianNeural", "de-DE-SeraphinaMultilingualNeural",
+            "de-DE-SeraphinaMultilingualNeural",
-            "de-DE-ConradNeural", "en-US-AndrewNeural", "en-US-AriaNeural",
+            "de-DE-AmalaNeural",
+            "de-DE-KatjaNeural",
+            "de-DE-KillianNeural",
+            "de-DE-ConradNeural",
+            "en-US-AriaNeural",
+            "en-US-JennyNeural",
         ])
 
         label("Whisper Modell")