diff --git a/wyoming_handler.py b/wyoming_handler.py
index 3d573a4..6a18d2f 100644
--- a/wyoming_handler.py
+++ b/wyoming_handler.py
@@ -3,7 +3,6 @@ import logging
 import time
 from typing import Dict, Optional
 
-import numpy as np
 from wyoming.audio import AudioChunk, AudioStart, AudioStop
 from wyoming.event import Event
 from wyoming.info import Describe, Info
@@ -94,10 +93,14 @@ class ChatterboxWyomingHandler(AsyncEventHandler):
             return True
 
         if Synthesize.is_type(event.type):
+            if self._streaming:
+                # Ignore duplicate Synthesize events sent alongside streaming protocol
+                return True
             synth = Synthesize.from_event(event)
             voice_name = synth.voice.name if synth.voice else None
             await self._synthesize_and_stream(synth.text, voice_name)
-            await self.write_event(SynthesizeStopped().event())
+            # NOTE: SynthesizeStopped is NOT sent here — it belongs only to the
+            # streaming protocol (SynthesizeStop path). Sending it here confuses HA.
             return True
 
         return True
@@ -134,27 +137,23 @@ class ChatterboxWyomingHandler(AsyncEventHandler):
                 logger.exception(f"Synthesis failed for chunk {i+1}")
                 continue
 
-            audio_np = audio_tensor.cpu().numpy().squeeze()
-            audio_int16 = (audio_np * 32767).clip(-32768, 32767).astype(np.int16)
+            audio_bytes = (
+                audio_tensor.cpu().numpy().squeeze() * 32767
+            ).astype("int16").tobytes()
 
             if first_chunk:
                 ttfa = time.monotonic() - start_time
                 logger.info(f"Time to first audio: {ttfa:.3f}s")
                 first_chunk = False
 
-            # Send in small sub-chunks so HA can begin playback immediately
-            # rather than waiting for the full audio blob to arrive.
-            audio_chunk_samples = 4096
-            for offset in range(0, len(audio_int16), audio_chunk_samples):
-                sub = audio_int16[offset : offset + audio_chunk_samples]
-                await self.write_event(
-                    AudioChunk(
-                        audio=sub.tobytes(),
-                        rate=sample_rate,
-                        width=2,
-                        channels=1,
-                    ).event()
-                )
+            await self.write_event(
+                AudioChunk(
+                    audio=audio_bytes,
+                    rate=sample_rate,
+                    width=2,
+                    channels=1,
+                ).event()
+            )
 
         await self.write_event(AudioStop().event())
         total = time.monotonic() - start_time