kd-engine 1.0.0__py3-none-any.whl

ytk/cleaning/service.py

@@ -0,0 +1,62 @@
+from ytk.cleaning.density import technical_density_score
+from ytk.cleaning.sentence_boundary import (
+    RuleBasedSentenceBoundaryBackend,
+    SentenceBoundaryBackend,
+)
+from ytk.models.content import ContentBlock, ImageRegionLocator, ParagraphLocator, TimeRangeLocator
+from ytk.models.transcript import TranscriptSegment
+
+
+class CleaningService:
+    def __init__(self, boundary_backend: SentenceBoundaryBackend | None = None) -> None:
+        self._boundary_backend = boundary_backend or RuleBasedSentenceBoundaryBackend()
+
+    def clean(self, segments: list[TranscriptSegment]) -> list[TranscriptSegment]:
+        return [self._boundary_backend.clean_segment(segment) for segment in segments]
+
+    def clean_blocks(self, blocks: list[ContentBlock] | tuple[ContentBlock, ...]) -> list[ContentBlock]:
+        """v2 block entry (T11.2): clean text while keeping block identity/locator.
+
+        Raw text is preserved; ``cleaned_text`` is a derived field only.  The
+        mapping stays strictly 1:1 so evidence can never drift to another block.
+        """
+        cleaned: list[ContentBlock] = []
+        for block in blocks:
+            probe = self._probe_segment(block)
+            normalized = self._boundary_backend.clean_segment(probe).cleaned_text
+            cleaned.append(block.model_copy(update={"cleaned_text": normalized}))
+        return cleaned
+
+    @staticmethod
+    def _probe_segment(block: ContentBlock) -> TranscriptSegment:
+        locator = block.locator
+        if isinstance(locator, TimeRangeLocator):
+            start, end = locator.time.start, locator.time.end
+        elif isinstance(locator, ParagraphLocator):
+            start = end = float(locator.paragraph)
+        elif isinstance(locator, ImageRegionLocator):
+            start = end = 0.0
+        else:  # video frame
+            start = end = float(getattr(locator, "time", 0.0))
+        return TranscriptSegment(
+            text=block.cleaned_text or block.raw_text,
+            raw_text=block.raw_text,
+            cleaned_text=None,
+            start=start,
+            end=end,
+            source="manual",
+        )
+
+    def _normalized_sentence(self, text: str) -> str:
+        probe = TranscriptSegment(
+            text=text,
+            raw_text=text,
+            cleaned_text=None,
+            start=0.0,
+            end=0.0,
+            source="manual",
+        )
+        return self._boundary_backend.clean_segment(probe).cleaned_text or ""
+
+    def score_sentence(self, text: str) -> float:
+        return technical_density_score(self._normalized_sentence(text))