diff --git a/Sumdays/app/src/main/java/com/example/sumdays/network/ApiClient.kt b/Sumdays/app/src/main/java/com/example/sumdays/network/ApiClient.kt
index 69ffc5f439..e744e8984e 100644
--- a/Sumdays/app/src/main/java/com/example/sumdays/network/ApiClient.kt
+++ b/Sumdays/app/src/main/java/com/example/sumdays/network/ApiClient.kt
@@ -7,7 +7,7 @@ import retrofit2.converter.gson.GsonConverterFactory
 import java.util.concurrent.TimeUnit
 
 object ApiClient {
-    const val BASE_URL = "http://172.30.1.51:3000/"
+    const val BASE_URL = "http://192.168.0.78:3000/"
     //폰에서 실행하려면 서버를 돌리는 컴퓨터의 IP를 여기 적어야 함
 
     // OkHttp 로깅 + 타임아웃 설정
diff --git a/sumdays-backend/ai/services/stt/stt_service.py b/sumdays-backend/ai/services/stt/stt_service.py
index dbaeedf0d2..0a5bb729d2 100644
--- a/sumdays-backend/ai/services/stt/stt_service.py
+++ b/sumdays-backend/ai/services/stt/stt_service.py
@@ -18,11 +18,19 @@ def transcribe(self, audio_file, language: str = "ko") -> str:
             model=self.model,
             file=buffer,
             language=language,
+            response_format="verbose_json",
         )
 
+        # check if file has no speech
+        if hasattr(result, 'segments') and result.segments:
+            avg_no_speech_prob = sum(s.no_speech_prob for s in result.segments) / len(result.segments)
+            if avg_no_speech_prob > 0.75:
+                return ""
+
+        # filter hallucination manually
         temp_filter = [
             "MBC 뉴스",
-            "시청해 주셔서",
+            "시청해",
             "구독",
             "고맙습니다."
         ]