| """Bruitage OCR/typo a longueur constante. |
| |
| Contrainte forte : les substitutions preservent la longueur du texte, pour que |
| les offsets gold restent valides sans re-mapping. C'est ce qui permet de scorer |
| les memes segments propres ET bruites (analyse §5.1, metrique robustesse OCR). |
| """ |
|
|
| import random |
| import unicodedata |
|
|
| |
| _OCR_SUBS = { |
| "0": "O", "O": "0", "o": "0", |
| "1": "l", "l": "1", "I": "1", |
| "5": "S", "S": "5", |
| "8": "B", "B": "8", |
| "é": "e", "è": "e", "ê": "e", "à": "a", "ç": "c", "û": "u", "ô": "o", |
| "É": "E", "È": "E", "À": "A", "Ç": "C", |
| "m": "n", "u": "v", |
| } |
|
|
|
|
| def _strip_accent(ch: str) -> str: |
| d = unicodedata.normalize("NFD", ch) |
| base = d[0] |
| return base if len(base) == 1 else ch |
|
|
|
|
| def apply_noise(text: str, rng: random.Random, rate: float = 0.03) -> str: |
| """Bruite ~rate des caracteres eligibles. Longueur strictement conservee.""" |
| chars = list(text) |
| for i, ch in enumerate(chars): |
| if rng.random() >= rate: |
| continue |
| r = rng.random() |
| if ch in _OCR_SUBS and r < 0.6: |
| chars[i] = _OCR_SUBS[ch] |
| elif ch.isalpha() and r < 0.8: |
| chars[i] = ch.upper() if ch.islower() else ch.lower() |
| elif ch.isalpha(): |
| chars[i] = _strip_accent(ch) |
| out = "".join(chars) |
| assert len(out) == len(text) |
| return out |
|
|
