Freeze independent DTW speech timing diagnostic and startup provenance

tools/speech-eval/test_dtw_alignment.py

@@ -0,0 +1,21 @@
+import unittest
+from dtw_alignment import aligned_segments
+
+
+class DtwTests(unittest.TestCase):
+    def raw(self):
+        return {"transcription": [{"text": " Hello. Again!", "offsets": {"from": 0, "to": 1000},
+            "tokens": [{"id": 100, "text": " Hello.", "t_dtw": 20, "offsets": {"from": 900, "to": 100}},
+                       {"id": 101, "text": " Again!", "t_dtw": 70}]}]}
+
+    def test_points_are_centiseconds_and_ordinary_reversal_not_repaired(self):
+        result = aligned_segments(self.raw(), 30000, 1000)
+        self.assertEqual([(r["start_ms"], r["end_ms"]) for r in result], [(30200, 30200), (30700, 30700)])
+        self.assertEqual("".join(r["text"] for r in result), " Hello. Again!")
+
+    def test_invalid_missing_fractional_and_nonmonotonic_points_rejected(self):
+        for point in [-1, 101, 1.5, None, 10]:
+            raw = self.raw()
+            raw["transcription"][0]["tokens"][1]["t_dtw"] = point
+            with self.assertRaises(ValueError):
+                aligned_segments(raw, 0, 1000)