Add voice activation (VAD) and audio fade

- Add webrtcvad dependency for real-time voice activity detection
- Create audio/fade.py with fade-in/fade-out utility
- Add VAD voice activation to client recording (sends audio only during speech)
- Apply 200ms fade-out to TTS output to avoid abrupt audio cuts
- Fix tts.py indentation error in except block

engine/tts.py

@@ -1,6 +1,7 @@
 from transformers import pipeline
 from config import Config
 import numpy as np
+from audio.fade import apply_fade
 
 
 class TTSEngine:
@@ -16,11 +17,11 @@ class TTSEngine:
                 device=0 if __import__("torch").cuda.is_available() else -1,
             )
         except Exception:
-       self.tts_pipeline = pipeline(
-            "text-to-speech",
-            model=self._tts_model,
-            device=-1,
-        )
+            self.tts_pipeline = pipeline(
+                "text-to-speech",
+                model=self._tts_model,
+                device=-1,
+            )
         self.tts_pipeline.start()
 
     def synthesize(self, text: str, output_sample_rate: int = 24000) -> np.ndarray:
@@ -50,4 +51,7 @@ class TTSEngine:
         
         audio = audio.astype(np.float32)
 
+        # Apply fade-out to avoid abrupt audio cuts
+        audio = apply_fade(audio, output_sample_rate, fade_duration_ms=200)
+
         return audio