PDF변환 추가

app/tasks.py

@@ -5,6 +5,7 @@ import os, json, subprocess, tempfile
 import httpx
 from celery import Celery
 from ocr_tasks import ocr_task  # noqa: F401
+from pdf_tasks import pdf_convert_task  # noqa: F401
 
 REDIS_URL      = os.getenv("REDIS_URL", "redis://redis:6379/0")
 MODEL_SIZE     = os.getenv("WHISPER_MODEL", "medium")
@@ -112,6 +113,22 @@ def _llm_call(prompt, model, use_openrouter, openrouter_url, openrouter_key, tim
                 f"해결: 설정에서 Ollama 타임아웃을 늘리거나, 더 작은 모델을 사용하세요."
             )
 
+def _extract_json_array(raw: str) -> str:
+    """LLM 응답에서 JSON 배열 문자열 추출"""
+    s = raw.strip()
+    if "```" in s:
+        parts = s.split("```")
+        # 홀수 인덱스 = 코드블록 내부
+        for part in parts[1::2]:
+            cleaned = part.lstrip("json").strip()
+            if cleaned.startswith("["):
+                return cleaned
+    # 코드블록 없을 때: [ ... ] 범위 직접 탐색
+    start, end = s.find("["), s.rfind("]")
+    if start != -1 and end > start:
+        return s[start:end+1]
+    return s
+
 def _translate_batch(texts, target_lang, use_or, model, or_url, or_key, timeout):
     if not texts or not model: return texts
     prompt = (
@@ -120,16 +137,17 @@ def _translate_batch(texts, target_lang, use_or, model, or_url, or_key, timeout)
         f"입력과 동일한 개수와 순서를 유지해.\n\n"
         f"{json.dumps(texts, ensure_ascii=False)}"
     )
-    try:
-        raw = _llm_call(prompt, model, use_or, or_url, or_key, timeout)
-        if "```" in raw: raw=raw.split("```")[1].lstrip("json\n").rstrip()
-        result = json.loads(raw)
-        if isinstance(result,list) and len(result)==len(texts):
-            return [str(r) for r in result]
-        return texts
-    except Exception as e:
-        print(f"[번역 실패] {e}")
-        return texts
+    raw = _llm_call(prompt, model, use_or, or_url, or_key, timeout)
+    json_str = _extract_json_array(raw)
+    result = json.loads(json_str)
+    if not isinstance(result, list):
+        raise Exception(f"번역 결과가 배열 형식이 아닙니다 (모델: {model})")
+    if len(result) != len(texts):
+        raise Exception(
+            f"번역 결과 개수 불일치: 입력 {len(texts)}개, 출력 {len(result)}개 "
+            f"(모델: {model}). 더 작은 청크 크기나 다른 모델을 시도하세요."
+        )
+    return [str(r) for r in result]
 
 def _refine_batch(texts, model, use_or, or_url, or_key, timeout):
     if not texts or not model: return texts
@@ -175,7 +193,7 @@ def _api_transcribe(audio_path, api_key, base_url, language, model="whisper-larg
     """Groq / OpenAI Whisper API 호출"""
     with open(audio_path,"rb") as f:
         data = f.read()
-    params = {"model":model}
+    params = {"model":model, "response_format":"verbose_json"}
     if language: params["language"] = language
     try:
         resp = httpx.post(
@@ -416,7 +434,10 @@ def subtitle_pipeline_task(
                 _prog(min(pct,95),3,f"{min(start+CHUNK,total)}/{total} 번역",
                       f"Step 3/3 — {_lang_name(translate_to)}로 번역 중... ({min(start+CHUNK,total)}/{total})")
                 batch=[s["text"] for s in chunk]
-                trans_texts.extend(_translate_batch(batch,translate_to,use_or,trans_model,openrouter_url,openrouter_key,timeout))
+                try:
+                    trans_texts.extend(_translate_batch(batch,translate_to,use_or,trans_model,openrouter_url,openrouter_key,timeout))
+                except Exception as e:
+                    raise Exception(f"LLM 번역 실패 ({_lang_name(translate_to)}, 청크 {ci+1}): {e}")
             translated_segments=[{**seg,"text":trans_texts[i] if i<len(trans_texts) else seg["text"]}
                                   for i,seg in enumerate(segments)]
             trans_suffix=translate_to