From 9d5594e848fc99f023f5fdf418d4934da4c9438b Mon Sep 17 00:00:00 2001
From: Yun Chan <yunchan@twentyoz.kr>
Date: Sun, 5 Apr 2026 10:16:10 +0900
Subject: [PATCH] =?UTF-8?q?STT=20VAD=20=EB=B9=88=20=EC=98=A4=EB=94=94?=
 =?UTF-8?q?=EC=98=A4=20=EC=97=90=EB=9F=AC=20=EB=B0=A9=EC=96=B4:=20VAD?=
 =?UTF-8?q?=EA=B0=80=20=EC=A0=84=EC=B2=B4=20=EC=A0=9C=EA=B1=B0=20=EC=8B=9C?=
 =?UTF-8?q?=20VAD=20=EC=97=86=EC=9D=B4=20=EC=9E=AC=EC=8B=9C=EB=8F=84?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- faster-whisper VAD가 짧은 오디오를 전부 무음 처리 → max() empty sequence 에러
- catch 후 vad_filter=False로 재시도하여 최소한 빈 텍스트 반환
---
 sidecar/main.py | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/sidecar/main.py b/sidecar/main.py
index 70051ca..78d9e51 100644
--- a/sidecar/main.py
+++ b/sidecar/main.py
@@ -222,7 +222,16 @@ async def transcribe(
             transcribe_kwargs["initial_prompt"] = initial_prompt
 
         # faster-whisper 전사 실행
-        segments_iter, info = _model.transcribe(audio_array, **transcribe_kwargs)
+        # VAD가 전체 오디오를 제거하면 max() 에러 발생 → VAD 없이 재시도
+        try:
+            segments_iter, info = _model.transcribe(audio_array, **transcribe_kwargs)
+        except ValueError as ve:
+            if "empty sequence" in str(ve) and transcribe_kwargs.get("vad_filter"):
+                logger.warning("VAD가 전체 오디오를 제거함 → VAD 없이 재시도")
+                transcribe_kwargs["vad_filter"] = False
+                segments_iter, info = _model.transcribe(audio_array, **transcribe_kwargs)
+            else:
+                raise
 
         # 세그먼트 수집
         segments_list: list[dict] = []