session.example.json

{
  "sources": [
    {
      "id": "readable",
      "path": "./readable.csv",
      "format": "csv",
      "roles": ["chunk", "text"],
      "columns": {
        "display_time": "発話時刻",
        "start": "発話時刻(秒)",
        "speaker": "話者名",
        "text": "書き起こし",
        "interviewee": "インタビュイー？"
      }
    },
    {
      "id": "diarization",
      "path": "./speaker-accurate.vtt",
      "format": "vtt",
      "roles": ["speaker", "timing"]
    },
    {
      "id": "context",
      "path": "./context-accurate.md",
      "format": "md",
      "roles": ["text"]
    }
  ],
  "alignment": {
    "skeleton_source": "readable",
    "boundary_source": "diarization",
    "min_anchor_chars": 4,
    "meaningful_speaker_score": 30,
    "low_margin_score": 15,
    "evidence_max_matches": 3
  },
  "speaker_map": {
    "Raw Moderator": "モデレーター",
    "Raw Interviewee": "ユーザー"
  },
  "fallback": {
    "metadata_field": "interviewee",
    "true_values": ["TRUE"],
    "true_speaker": "ユーザー",
    "false_speaker": "モデレーター",
    "default_speaker": "モデレーター"
  },
  "replacements": [
    {"from": "誤認識された固有語", "to": "確認済みの固有語"}
  ],
  "overrides": [
    {
      "segment": 10,
      "text": "文脈と複数ソースから確認した本文です。"
    },
    {
      "segment": 11,
      "speaker": "ユーザー"
    },
    {
      "segment": 12,
      "chunks": [
        {"speaker": "モデレーター", "text": "質問の本文です。"},
        {"speaker": "ユーザー", "text": "回答の本文です。"}
      ]
    }
  ],
  "output": {
    "path": "./out/draft.md",
    "review_path": "./out/review.jsonl",
    "allowed_speakers": ["モデレーター", "ユーザー"],
    "label_template": "**{speaker}:** {text}",
    "label_regex": "^\\*\\*(?P<speaker>.+?):\\*\\*\\s+(?P<text>.+)$",
    "prohibit_headings": true,
    "forbidden_patterns": ["SPEAKER_[0-9]+", "\\b(?:TRUE|FALSE)\\b"],
    "quote_pairs": [["「", "」"], ["『", "』"]]
  }
}