fix(analyze): extract text from Whisper JSON format

- Handle Whisper transcription format (array of segments with text)
- Extract full text from transcription array before passing to DeepAnalyzer
- Fix offsets parsing to support both dict and list formats

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

kb_pipeline/processors/analyze.py

@@ -81,6 +81,15 @@ class AnalyzeProcessor(BaseProcessor):
             metadata = self._load_json(input_data.metadata_file)
             content = self._load_json(input_data.content_file)
 
+            # 处理 Whisper 字幕格式 - 提取完整文本
+            if content and 'transcription' in content:
+                # Whisper JSON 格式，提取文本
+                text_parts = []
+                for item in content.get('transcription', []):
+                    text_parts.append(item.get('text', ''))
+                content = {'text': ' '.join(text_parts)}
+                logger.info(f"Extracted {len(content['text'])} chars from Whisper transcription")
+
             # 对于文章类型，内容可能是纯文本文件
             if not content and input_data.content_type in ("article", "answer", "zhuanlan"):
                 content_path = Path(input_data.content_file)

kb_pipeline/processors/asr.py

@@ -167,6 +167,92 @@ class ASRProcessor:
                     self.state_manager.save(state)
             raise
 
+    def _check_existing_subtitle_files(self, input_data: ASRInput) -> Optional[ASROutput]:
+        """检查是否已存在字幕文件（幂等性检查）
+
+        Args:
+            input_data: ASR 输入数据
+
+        Returns:
+            如果找到现有字幕文件返回 ASROutput，否则返回 None
+        """
+        output_dir = Path(input_data.content_dir)
+
+        # 尝试两种命名格式：新格式（带 _subtitle 后缀）和旧格式
+        base_names = [
+            f"{input_data.platform}_{input_data.content_id}_subtitle",
+            f"{input_data.platform}_{input_data.content_id}",
+        ]
+
+        for base_name in base_names:
+            json_path = output_dir / f"{base_name}.json"
+            srt_path = output_dir / f"{base_name}.srt"
+            txt_path = output_dir / f"{base_name}.txt"
+
+            if json_path.exists():
+                # 读取现有字幕数据
+                segments = []
+                text = ""
+                duration = 0.0
+
+                try:
+                    with open(json_path, 'r', encoding='utf-8') as f:
+                        data = json.load(f)
+
+                    # 解析字幕格式
+                    if 'transcription' in data:
+                        # Whisper 格式
+                        for i, item in enumerate(data['transcription']):
+                            offsets = item.get('offsets', {})
+                            if isinstance(offsets, dict):
+                                # 新格式：{"from": 0, "to": 5320}
+                                start = offsets.get('from', 0) / 1000.0
+                                end = offsets.get('to', 0) / 1000.0
+                            else:
+                                # 旧格式：[0, 5320]
+                                start = offsets[0] / 1000.0 if len(offsets) > 0 else 0
+                                end = offsets[1] / 1000.0 if len(offsets) > 1 else 0
+                            segments.append({
+                                'id': i,
+                                'start': start,
+                                'end': end,
+                                'text': item.get('text', '')
+                            })
+                    elif 'body' in data:
+                        # B站官方字幕格式
+                        for i, item in enumerate(data['body']):
+                            segments.append({
+                                'id': i,
+                                'start': item.get('from', 0),
+                                'end': item.get('to', 0),
+                                'text': item.get('content', '')
+                            })
+
+                    text = ' '.join(s['text'] for s in segments)
+                    if segments:
+                        duration = segments[-1]['end']
+
+                    output_files = {'json': str(json_path)}
+                    if srt_path.exists():
+                        output_files['srt'] = str(srt_path)
+                    if txt_path.exists():
+                        output_files['txt'] = str(txt_path)
+
+                    return ASROutput(
+                        success=True,
+                        tool="existing",
+                        has_subtitle=True,
+                        output_files=output_files,
+                        segments=segments,
+                        text=text,
+                        duration_seconds=duration
+                    )
+                except Exception as e:
+                    logger.warning(f"Failed to read existing subtitle file: {e}")
+                    continue
+
+        return None
+
     async def _try_official_subtitle(self, input_data: ASRInput) -> Optional[ASROutput]:
         """尝试下载 B站官方字幕
 
@@ -433,10 +519,19 @@ class ASRProcessor:
                 # Whisper.cpp JSON 格式
                 if 'transcription' in data:
                     for i, item in enumerate(data['transcription']):
+                        offsets = item.get('offsets', {})
+                        if isinstance(offsets, dict):
+                            # 新格式：{"from": 0, "to": 5320}
+                            start = offsets.get('from', 0) / 1000.0
+                            end = offsets.get('to', 0) / 1000.0
+                        else:
+                            # 旧格式：[0, 5320]
+                            start = offsets[0] / 1000.0 if len(offsets) > 0 else 0
+                            end = offsets[1] / 1000.0 if len(offsets) > 1 else 0
                         segments.append({
                             'id': i,
-                            'start': item.get('offsets', [0, 0])[0] / 1000.0,
-                            'end': item.get('offsets', [0, 0])[1] / 1000.0,
+                            'start': start,
+                            'end': end,
                             'text': item.get('text', '')
                         })
                 text = ' '.join(s['text'] for s in segments)