fixed a bug

sentence_api/tests/test_processing.py

@@ -323,3 +323,61 @@ def test_reprocess_without_moss_clears_original_segments(tmp_path, monkeypatch):
     VideoProcessor(settings, repository, UnavailableTranscriber()).process(VIDEO_HASH)
     assert repository.get_moss_segments(VIDEO_HASH) == []
 
+
+def test_mark_processing_clears_previous_sentence_data(tmp_path):
+    repository = VideoRepository(tmp_path / "oral.sqlite3")
+    repository.upsert_upload(
+        video_hash=VIDEO_HASH,
+        title="MOSS lesson",
+        filename="lesson.mp4",
+        stored_filename=f"{VIDEO_HASH}.mp4",
+        content_type="video/mp4",
+        size_bytes=5,
+        language="en",
+    )
+    repository.save_processing_result(
+        SentenceBoundaryDocument(
+            video_hash=VIDEO_HASH,
+            duration_ms=2400,
+            algorithm_version=MOSS_ALGORITHM_VERSION,
+            sentences=[
+                SentenceBoundary(
+                    index=0,
+                    start_ms=0,
+                    end_ms=1200,
+                    text="Old first.",
+                    language="en",
+                    reference_speech_duration_ms=1000,
+                ),
+                SentenceBoundary(
+                    index=1,
+                    start_ms=1200,
+                    end_ms=2400,
+                    text="Old second.",
+                    language="en",
+                    reference_speech_duration_ms=1000,
+                ),
+            ],
+        ),
+        "Old first. Old second.",
+        moss_segments=[
+            {
+                "start_seconds": 0.0,
+                "end_seconds": 1.2,
+                "text": "Old first.",
+                "speaker": "S01",
+            }
+        ],
+    )
+
+    repository.mark_processing(VIDEO_HASH)
+
+    video = repository.get_video(VIDEO_HASH)
+    document = repository.get_document(VIDEO_HASH)
+    assert video["status"] == "processing"
+    assert video["transcription"] is None
+    assert video["sentence_count"] == 0
+    assert document is not None
+    assert document.sentences == []
+    assert repository.get_moss_segments(VIDEO_HASH) == []
+

sentence_api/tests/test_transcription.py

@@ -423,6 +423,36 @@ def test_split_sentences_ignores_am_pm_abbreviations():
     ]
 
 
+def test_split_sentences_ignores_clock_times_with_pm_period():
+    compact = TranscriptionSegment(
+        start_seconds=0.0,
+        end_seconds=6.0,
+        text="The briefing is at 3:00 PM. Eastern officials will attend later.",
+    )
+    assert [s.text for s in split_sentences_at_punctuation([compact])] == [
+        "The briefing is at 3:00 PM. Eastern officials will attend later."
+    ]
+
+    spoken = TranscriptionSegment(
+        start_seconds=0.0,
+        end_seconds=6.0,
+        text="We start at 5 pm. Eastern time and then break.",
+    )
+    assert [s.text for s in split_sentences_at_punctuation([spoken])] == [
+        "We start at 5 pm. Eastern time and then break."
+    ]
+
+    morning = TranscriptionSegment(
+        start_seconds=0.0,
+        end_seconds=6.0,
+        text="It begins at 9:30 AM. Please arrive early.",
+    )
+    assert [s.text for s in split_sentences_at_punctuation([morning])] == [
+        "It begins at 9:30 AM.",
+        "Please arrive early.",
+    ]
+
+
 def test_split_sentences_merges_segment_ending_with_abbreviation():
     segments = [
         TranscriptionSegment(0.0, 5.0, "The meeting ends at 5 p.m."),
@@ -460,6 +490,77 @@ def test_split_sentences_ignores_common_abbreviations():
     ]
 
 
+def test_split_sentences_merges_same_speaker_line_wraps():
+    segments = [
+        TranscriptionSegment(
+            11.18,
+            15.45,
+            "[S01]Let's begin with some scheduling announcements for the remainder of President Trump's very",
+        ),
+        TranscriptionSegment(15.45, 16.29, "[S01]busy week."),
+        TranscriptionSegment(
+            16.29,
+            21.8,
+            "[S01]Today at 3:00 PM, President Trump will host a Black History Month reception in the East",
+        ),
+        TranscriptionSegment(21.8, 23.29, "[S01]Room of the White House."),
+    ]
+    sentences = split_sentences_at_punctuation(segments)
+    assert [s.text for s in sentences] == [
+        (
+            "Let's begin with some scheduling announcements for the remainder of "
+            "President Trump's very busy week."
+        ),
+        (
+            "Today at 3:00 PM, President Trump will host a Black History Month "
+            "reception in the East Room of the White House."
+        ),
+    ]
+    assert sentences[0].start_seconds == 11.18
+    assert sentences[0].end_seconds == 16.29
+    assert sentences[1].start_seconds == 16.29
+    assert sentences[1].end_seconds == 23.29
+
+
+def test_split_sentences_splits_when_speaker_changes():
+    segments = [
+        TranscriptionSegment(4.63, 5.93, "[S01]Hello, everybody."),
+        TranscriptionSegment(6.0, 6.53, "[S02]Hello."),
+        TranscriptionSegment(6.56, 7.83, "[S01]Good afternoon."),
+    ]
+    sentences = split_sentences_at_punctuation(segments)
+    assert [s.text for s in sentences] == [
+        "Hello, everybody.",
+        "Hello.",
+        "Good afternoon.",
+    ]
+
+
+def test_split_sentences_keeps_same_speaker_sentences_apart():
+    segments = [
+        TranscriptionSegment(6.56, 7.83, "[S01]Good afternoon."),
+        TranscriptionSegment(7.93, 9.86, "[S01]Thank you for your patience, as always."),
+    ]
+    sentences = split_sentences_at_punctuation(segments)
+    assert [s.text for s in sentences] == [
+        "Good afternoon.",
+        "Thank you for your patience, as always.",
+    ]
+
+
+def test_split_sentences_ignores_initials_and_version_numbers():
+    segment = TranscriptionSegment(
+        start_seconds=0.0,
+        end_seconds=8.0,
+        text="Donald J. Trump discussed version 1.9 today. Then the meeting ended.",
+    )
+    sentences = split_sentences_at_punctuation([segment])
+    assert [s.text for s in sentences] == [
+        "Donald J. Trump discussed version 1.9 today.",
+        "Then the meeting ended.",
+    ]
+
+
 def test_post_audio_requests_word_timestamps_and_falls_back(monkeypatch):
     calls = []