[PratikDhanave]

Summary

The English number normalizer never normalizes the ordinal "ninth", and breaks every compound ordinal ending in nine.

Problem

EnglishNumberNormalizer.ones_ordinal (whisper/normalizers/english.py) is generated by appending "th" to each cardinal, special-casing only words ending in "t":

name + ("h" if name.endswith("t") else "th")   # eight -> eighth

"nine" does not end in "t", so this produces the non-word key "nineth". The real ordinal "ninth" is therefore absent from the map, so it is left un-normalized:

input	current	expected
ninth	ninth	9th
the ninth of may	the ninth of may	the 9th of may
twenty ninth	20 ninth	29th
ninety ninth	90 ninth	99th
one hundred ninth	100 ninth	109th

("nineth" — a string that never appears in real text — is present in the map, which confirms the generator simply targeted the wrong key.)

Fix

Special-case "ninth" explicitly, exactly like the other irregular ordinals (first/second/third/fifth/twelfth), and exclude 9 from the generated set:

             "fifth": (5, "th"),
+            "ninth": (9, "th"),
             "twelfth": (12, "th"),
             **{
                 name + ("h" if name.endswith("t") else "th"): (value, "th")
                 for name, value in self.ones.items()
-                if value > 3 and value != 5 and value != 12
+                if value > 3 and value not in (5, 9, 12)
             },

Validation

Added ordinal assertions to tests/test_normalizer.py (ninth, twenty ninth, ninety ninth, one hundred ninth).

• Before the fix: assert std("ninth") == "9th" fails ('ninth' != '9th').
• After the fix: all pass.
• Full tests/test_normalizer.py passes (4 passed) with the change; the existing suite had no ordinal-ending-in-nine coverage, which is why this shipped.

---

Developed with AI assistance: the bug was located, the fix applied, and tests written and run locally (pytest). I have reviewed and verified the change and take responsibility for it.

[koksalenes]

Correctly handles ordinal numbers and plural forms now