add merging sentences action

sentence_api/repository.py

@@ -548,6 +548,115 @@ class VideoRepository:
             for row in result_rows
         ]
 
+
+    def merge_with_next_sentence(
+        self,
+        video_hash: str,
+        sentence_index: int,
+    ) -> Optional[SentenceBoundary]:
+        """Merge sentence_index with the immediately following sentence."""
+        normalized_hash = normalize_video_hash(video_hash)
+        with self._connect() as connection:
+            current = connection.execute(
+                """
+                SELECT * FROM sentences
+                WHERE video_hash = ? AND sentence_index = ?
+                """,
+                (normalized_hash, sentence_index),
+            ).fetchone()
+            if current is None:
+                return None
+
+            nxt = connection.execute(
+                """
+                SELECT * FROM sentences
+                WHERE video_hash = ? AND sentence_index = ?
+                """,
+                (normalized_hash, sentence_index + 1),
+            ).fetchone()
+            if nxt is None:
+                raise ValueError("No next sentence to merge with.")
+
+            merged_text = _join_adjacent_sentence_text(
+                current["text"] or "",
+                nxt["text"] or "",
+                current["language"] or nxt["language"],
+            )
+            merged_language = current["language"] or nxt["language"]
+            merged_speech_duration = max(
+                1,
+                int(current["reference_speech_duration_ms"] or 0)
+                + int(nxt["reference_speech_duration_ms"] or 0),
+            )
+            if nxt["end_ms"] <= current["start_ms"]:
+                raise ValueError("Merged sentence times must increase.")
+
+            connection.execute(
+                """
+                UPDATE sentences
+                SET end_ms = ?, text = ?, language = ?,
+                    reference_speech_duration_ms = ?
+                WHERE video_hash = ? AND sentence_index = ?
+                """,
+                (
+                    nxt["end_ms"],
+                    merged_text,
+                    merged_language,
+                    merged_speech_duration,
+                    normalized_hash,
+                    sentence_index,
+                ),
+            )
+            connection.execute(
+                """
+                DELETE FROM sentences
+                WHERE video_hash = ? AND sentence_index = ?
+                """,
+                (normalized_hash, sentence_index + 1),
+            )
+
+            # Compact following indexes upward so sentence_index stays contiguous.
+            following = [
+                row["sentence_index"]
+                for row in connection.execute(
+                    """
+                    SELECT sentence_index FROM sentences
+                    WHERE video_hash = ? AND sentence_index > ?
+                    ORDER BY sentence_index ASC
+                    """,
+                    (normalized_hash, sentence_index + 1),
+                ).fetchall()
+            ]
+            for old_index in following:
+                connection.execute(
+                    """
+                    UPDATE sentences SET sentence_index = ?
+                    WHERE video_hash = ? AND sentence_index = ?
+                    """,
+                    (old_index - 1, normalized_hash, old_index),
+                )
+
+            connection.execute(
+                "UPDATE videos SET updated_at = ? WHERE video_hash = ?",
+                (utc_now(), normalized_hash),
+            )
+            row = connection.execute(
+                """
+                SELECT * FROM sentences
+                WHERE video_hash = ? AND sentence_index = ?
+                """,
+                (normalized_hash, sentence_index),
+            ).fetchone()
+
+        return SentenceBoundary(
+            index=row["sentence_index"],
+            start_ms=row["start_ms"],
+            end_ms=row["end_ms"],
+            text=row["text"],
+            language=row["language"],
+            reference_speech_duration_ms=row["reference_speech_duration_ms"],
+        )
+
     def adjust_sentence_boundary(
         self,
         video_hash: str,
@@ -1019,3 +1128,40 @@ class VideoRepository:
                 (normalize_video_hash(video_hash),),
             )
         return video
+
+
+def _join_adjacent_sentence_text(
+    left: str,
+    right: str,
+    language: Optional[str],
+) -> str:
+    """Join two adjacent sentence texts with language-aware spacing."""
+    left_text = left.strip()
+    right_text = right.strip()
+    if not left_text:
+        return right_text
+    if not right_text:
+        return left_text
+
+    lang = (language or "").strip().lower()
+    if lang.startswith(("zh", "ja", "ko")):
+        return f"{left_text}{right_text}"
+
+    left_end = left_text[-1]
+    right_start = right_text[0]
+    if _is_cjk_char(left_end) or _is_cjk_char(right_start):
+        return f"{left_text}{right_text}"
+    if left_end.isspace() or right_start in ",.;:!?)]}、，。！？；：”’》〉】":
+        return f"{left_text}{right_text}"
+    return f"{left_text} {right_text}"
+
+
+def _is_cjk_char(char: str) -> bool:
+    code = ord(char)
+    return (
+        0x3400 <= code <= 0x4DBF
+        or 0x4E00 <= code <= 0x9FFF
+        or 0xF900 <= code <= 0xFAFF
+        or 0x20000 <= code <= 0x2CEAF
+    )
+