60 lines
2.8 KiB
Python
60 lines
2.8 KiB
Python
|
"""Structural tests only; actual word-time quality requires genuine inference."""
|
||
|
|
import unittest
|
||
|
|
from token_alignment import aligned_segments
|
||
|
|
|
||
|
|
|
||
|
|
def fixture():
|
||
|
|
tokens = [{"id": 100 + index, "text": text,
|
||
|
|
"offsets": {"from": start, "to": end}}
|
||
|
|
for index, (text, start, end) in enumerate([
|
||
|
|
(" Hello", 100, 400), (" world.", 400, 900),
|
||
|
|
(" Guten", 1100, 1400), (" Tag!", 1400, 1900)])]
|
||
|
|
return {"transcription": [{"text": " Hello world. Guten Tag!",
|
||
|
|
"offsets": {"from": 0, "to": 2000}, "tokens": tokens}]}
|
||
|
|
|
||
|
|
|
||
|
|
class TokenAlignmentTests(unittest.TestCase):
|
||
|
|
def test_exact_sentences_absolute_nonzero_source_coordinates(self):
|
||
|
|
result = aligned_segments(fixture(), 30000, 2000)
|
||
|
|
self.assertEqual([row["text"] for row in result], [" Hello world.", " Guten Tag!"])
|
||
|
|
self.assertEqual([(row["start_ms"], row["end_ms"]) for row in result],
|
||
|
|
[(30100, 30900), (31100, 31900)])
|
||
|
|
self.assertEqual(result[0]["original_start_ms"], 30000)
|
||
|
|
|
||
|
|
def test_structural_tokens_never_enter_candidates(self):
|
||
|
|
raw = fixture()
|
||
|
|
raw["transcription"][0]["tokens"].insert(0, {"id": 50364, "text": "[_BEG_]"})
|
||
|
|
self.assertEqual(len(aligned_segments(raw, 0, 2000)), 2)
|
||
|
|
|
||
|
|
def test_missing_nonmonotonic_outside_segment_and_invalid_special_rejected(self):
|
||
|
|
mutations = [lambda token: token.pop("offsets"),
|
||
|
|
lambda token: token.update(offsets={"from": -1, "to": 100}),
|
||
|
|
lambda token: token.update(offsets={"from": 300, "to": 100}),
|
||
|
|
lambda token: token.update(offsets={"from": 0, "to": 2500}),
|
||
|
|
lambda token: token.update(id=50364, text="ordinary words"),
|
||
|
|
lambda token: token.update(text=" changed")]
|
||
|
|
for mutate in mutations:
|
||
|
|
raw = fixture()
|
||
|
|
mutate(raw["transcription"][0]["tokens"][1])
|
||
|
|
with self.assertRaises(ValueError):
|
||
|
|
aligned_segments(raw, 0, 2000)
|
||
|
|
|
||
|
|
def test_no_punctuation_preserves_whole_text(self):
|
||
|
|
raw = fixture()
|
||
|
|
for token in raw["transcription"][0]["tokens"]:
|
||
|
|
token["text"] = token["text"].rstrip(".!")
|
||
|
|
raw["transcription"][0]["text"] = "".join(t["text"] for t in raw["transcription"][0]["tokens"])
|
||
|
|
result = aligned_segments(raw, 0, 2000)
|
||
|
|
self.assertEqual(len(result), 1)
|
||
|
|
self.assertEqual(result[0]["text"], raw["transcription"][0]["text"])
|
||
|
|
|
||
|
|
def test_empty_speech_is_valid_and_missing_tokens_are_rejected(self):
|
||
|
|
self.assertEqual(aligned_segments({"transcription": []}, 0, 1000), [])
|
||
|
|
raw = fixture()
|
||
|
|
del raw["transcription"][0]["tokens"]
|
||
|
|
with self.assertRaises(ValueError):
|
||
|
|
aligned_segments(raw, 0, 2000)
|
||
|
|
|
||
|
|
|
||
|
|
if __name__ == "__main__":
|
||
|
|
unittest.main()
|