fixed a bug

sentence_api/tests/test_processing.py

@@ -5,6 +5,7 @@ from sentence_api.processing import (
     VideoProcessor,
     document_from_transcript,
     find_sentence_gaps,
+    _plan_gap_chunks,
     split_sentence_text,
 )
 from sentence_api.models import SentenceBoundary, SentenceBoundaryDocument
@@ -417,6 +418,22 @@ def test_find_sentence_gaps_ignores_short_pauses():
     assert find_sentence_gaps(document) == []
 
 
+def test_plan_gap_chunks_splits_long_gap_at_silence():
+    samples = np.concatenate(
+        [_speech(14.0), _silence(0.6), _speech(16.0)]
+    ).astype(np.float32)
+    chunks = _plan_gap_chunks(samples, SAMPLE_RATE, int(samples.size / SAMPLE_RATE * 1000))
+
+    assert len(chunks) == 2
+    first_start, first_end = chunks[0]
+    second_start, second_end = chunks[1]
+    assert first_start == 0
+    assert 13_000 <= first_end <= 17_000
+    assert second_start == first_end
+    assert second_end == int(samples.size / SAMPLE_RATE * 1000)
+    assert all(end - start <= 20_000 for start, end in chunks)
+
+
 def test_repair_gaps_inserts_transcribed_sentences_and_keeps_existing(tmp_path, monkeypatch):
     from dataclasses import replace