diff --git a/bin/api/voice.py b/bin/api/voice.py
index 8be5a49..68c1eb1 100644
--- a/bin/api/voice.py
+++ b/bin/api/voice.py
@@ -5,8 +5,6 @@ import json
 import uuid
 from pathlib import Path
 
-from bin.commands.protocols import EventType, MsgType, full_client_request, receive_message
-
 _RESOURCE_ID = "volc.service_type.10029"
 _ENDPOINT = "wss://openspeech.bytedance.com/api/v3/tts/unidirectional/stream"
 
@@ -33,6 +31,9 @@ def synthesize_tts(
 
     Returns absolute path to the output audio file.
     """
+    # Delayed import to avoid circular dependency when bin.commands is not yet loaded
+    from bin.commands.protocols import EventType, MsgType, full_client_request, receive_message
+
     try:
         import websockets
     except ImportError as e: