diff --git a/src/main.py b/src/main.py
index 73285ef..5e2952f 100644
--- a/src/main.py
+++ b/src/main.py
@@ -12,19 +12,65 @@ client = TritonGrpcClient(TRITON_URL)
 
 @app.websocket("/asr")
 async def websocket_endpoint(ws: WebSocket):
+    print("Starting ASR2 WebSocket")
+
+    try:
+        await client.websocket_stream_from_websocket(ws, filename="stream.webm", accept_in_client=True)
+
+    except Exception as e:
+        print("ASR2 WebSocket exception:", e)
+        try:
+            await ws.close()
+        except:
+            pass
+
+
+
+@app.websocket("/asr_3")
+async def websocket_endpoint(ws: WebSocket):
+    """
+    WebSocket endpoint for real-time ASR.
+    Accumulates audio and processes only once when complete.
+    """
     print("Starting ASR service")
     await ws.accept()
 
     audio_buffer = bytearray()
 
     try:
+        # Receive all audio chunks
         while True:
-            data = await ws.receive_bytes()
-            audio_buffer.extend(data)
+            try:
+                # Try to receive bytes
+                data = await ws.receive_bytes()
+                audio_buffer.extend(data)
+            except:
+                # Try to receive text (for EOF signal)
+                message = await ws.receive_text()
+                msg_data = json.loads(message)
 
-            async for s in client.event_stream_from_bytes(raw=bytes(audio_buffer), filename="stream.webm"):
+                # Check for EOF signal
+                if msg_data.get("eof"):
+                    print("Received EOF signal, processing audio...")
+                    break
+
+        # Process complete audio buffer ONCE
+        if len(audio_buffer) > 0:
+            print(f"Processing {len(audio_buffer)} bytes of audio")
+            async for s in client.event_stream_from_bytes(
+                raw=bytes(audio_buffer),
+                filename="stream.webm"
+            ):
                 await ws.send_text(json.dumps(s, ensure_ascii=False))
 
     except Exception as e:
-        print("WebSocket error:", e)
-        await ws.close()
+        print(f"WebSocket error: {e}")
+        await ws.send_text(json.dumps({
+            "event": "error",
+            "detail": str(e)
+        }, ensure_ascii=False))
+    finally:
+        try:
+            await ws.close()
+        except:
+            pass
diff --git a/test/test.py b/test/test.py
index f395142..3f619e4 100644
--- a/test/test.py
+++ b/test/test.py
@@ -3,28 +3,39 @@ import websockets
 import json
 
 AUDIO_FILE = "./data/khabar_1_5.wav"
-WS_URL = "ws://127.0.0.1:7200/asr"
+WS_URL = "ws://127.0.0.1:3002/asr"
 
 async def send_audio():
-    async with websockets.connect(WS_URL) as ws:
+    async with websockets.connect(WS_URL, max_size=None) as ws:
         print("Connected to ASR WebSocket")
 
-        # Send audio in chunks
+        # --- Send audio in binary chunks ---
         with open(AUDIO_FILE, "rb") as f:
             chunk_size = 100000
             while chunk := f.read(chunk_size):
                 await ws.send(chunk)
 
-        # Tell server no more audio is coming
-        await ws.send(json.dumps({"eof": True}))
+        # --- Tell server that audio stream is finished ---
+        await ws.send(json.dumps({"event": "end"}))
 
-        # Receive results
+        # --- Receive partial + final transcripts ---
         try:
             while True:
-                message = await ws.recv()
-                data = json.loads(message)
+                msg = await ws.recv()
+
+                # Triton code may send JSON or text
+                try:
+                    data = json.loads(msg)
+                except Exception:
+                    print("Non-JSON frame:", msg)
+                    continue
+
                 print("Received:", data)
 
+                # Optional: stop when server sends is_final
+                if data.get("is_final"):
+                    print("Final:", data["text"])
+
         except websockets.exceptions.ConnectionClosed:
             print("WebSocket closed by server")