feat: align_segments() — map Whisper timestamps to pyannote speakers

2026-04-02 01:00:58 +02:00
parent 1a9d0eacc2
commit b8cc8a3b33
2 changed files with 65 additions and 0 deletions
@@ -0,0 +1,33 @@
+def test_align_assigns_speaker_by_overlap():
+    from alignment import align_segments
+    whisper = [
+        {"start": 0.0, "end": 2.0, "text": "Hallo"},
+        {"start": 2.1, "end": 4.0, "text": "Wie geht es"},
+    ]
+    speakers = [
+        (0.0, 2.5, "SPEAKER_00"),
+        (2.5, 5.0, "SPEAKER_01"),
+    ]
+    result = align_segments(whisper, speakers)
+    assert result[0] == ("SPEAKER_00", "Hallo")
+    assert result[1] == ("SPEAKER_01", "Wie geht es")
+
+
+def test_align_merges_consecutive_same_speaker():
+    from alignment import align_segments
+    whisper = [
+        {"start": 0.0, "end": 1.0, "text": "Hallo"},
+        {"start": 1.1, "end": 2.0, "text": "Welt"},
+    ]
+    speakers = [(0.0, 3.0, "SPEAKER_00")]
+    result = align_segments(whisper, speakers)
+    assert len(result) == 1
+    assert result[0] == ("SPEAKER_00", "Hallo Welt")
+
+
+def test_align_fallback_when_no_speaker_overlap():
+    from alignment import align_segments
+    whisper = [{"start": 0.0, "end": 1.0, "text": "Hallo"}]
+    speakers = []
+    result = align_segments(whisper, speakers)
+    assert result[0][0] == "SPEAKER_00"