french was never going to work, and a test said it would

Probing all nine of Kokoro's languages through the running endpoint, one request
each: eight returned audio and `fr` returned a 500 — "espeak-ng has no such
language". French is regioned in espeak exactly like English (fr-fr, fr-be,
fr-ch), so the bare code every OpenAI-compatible client sends was refused. The
French voice had never worked.

The regionless map's own comment asserted the opposite: that "it", "fr" and "de"
are all accepted bare. Two of those are true. A test asserted the same thing for
all three, so the belief was pinned in place without ever being measured —
exactly the failure the Kokoro-English fix was about, one language further along.
"fr" now maps to "fr-fr"; the comment says not to add a code on the strength of
the pattern, because the pattern is what got this wrong.

Measured bare-OK while I was there, so the next person has data instead of a
guess: de, it, nl, pl, tr, ru, ko, ar, cs, hu, sv. The corrected test asserts
those, and a new one checks the invariant that broke here — every language the
listing advertises for Kokoro must resolve to something espeak was measured to
accept.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014T3rZozi1Ehk96NG1WesgC

codai/__init__.py

@@ -16,7 +16,7 @@
 
 # Canonical product version for CoderAI — single source of truth. Both the API
 # metadata and the admin web UI read from here.
-__version__ = "0.5.39"
+__version__ = "0.5.40"
 __tagline__ = "Complete orchestration, distribution and escalation of remotizable advanced inference."
 
 # Configure the CUDA caching allocator BEFORE torch is imported anywhere.

codai/api/tts_backends.py

@@ -237,10 +237,17 @@ def _kokoro_static_voices(config: Optional[dict] = None) -> list:
 #: Every OpenAI-compatible client in existence sends "en", so Kokoro's English
 #: voices were unreachable while its Italian ones worked, which reads as the
 #: model being broken rather than as a two-letter code needing a region.
-#: Only codes espeak actually refuses belong here: "it", "fr" and "de" are
-#: accepted bare and must pass through untouched.
+#: Only codes espeak actually refuses belong here. Measured against the running
+#: engine 2026-10-30, one language per request: "de", "it", "nl", "pl", "tr",
+#: "ru", "ko", "ar", "cs", "hu" and "sv" are all accepted bare and must pass
+#: through untouched -- but "fr" is NOT, which this comment previously asserted
+#: it was. French is regioned in espeak exactly like English (fr-fr, fr-be,
+#: fr-ch), so Kokoro's French voice answered every request with a 500 while its
+#: Italian one worked. Do not add a code here on the strength of the pattern:
+#: probe it, because the pattern is what got "fr" wrong.
 _ESPEAK_REGIONLESS = {
     "en": "en-us",
+    "fr": "fr-fr",
     "pt": "pt-br",
     "zh": "cmn",
 }

tests/test_full_pass_regressions.py

@@ -128,12 +128,26 @@ def test_a_bare_en_becomes_an_espeak_language():
 
 
 def test_languages_espeak_accepts_bare_are_untouched():
-    """Rewriting 'it' would break the voices that were working."""
+    """Rewriting 'it' would break the voices that were working.
+
+    'fr' USED to be in this list, on the strength of the pattern rather than a
+    measurement, and it was wrong: probing the running engine on 2026-10-30
+    showed French is regioned in espeak exactly like English, so a bare 'fr'
+    answered every request with a 500 while 'it' worked. It now maps to 'fr-fr'
+    and lives in the mapped set below. The rest of these were measured bare.
+    """
     from codai.api.tts_backends import espeak_language
-    for code in ("it", "fr", "de", "es", "en-gb", "en-us"):
+    for code in ("it", "de", "es", "nl", "pl", "tr", "ru", "ko", "ar",
+                 "cs", "hu", "sv", "en-gb", "en-us"):
         assert espeak_language(code) == code
 
 
+def test_french_is_regioned_like_english():
+    from codai.api.tts_backends import espeak_language
+    assert espeak_language("fr") == "fr-fr"
+    assert espeak_language("fr-be") == "fr-be"
+
+
 def test_an_empty_language_still_has_a_default():
     from codai.api.tts_backends import espeak_language
     assert espeak_language("") == "en-us"

tests/test_tts_languages_are_published.py

@@ -305,3 +305,64 @@ def test_xtts_set_matches_the_checkpoints_own_config():
     if not declared:
         pytest.skip('this XTTS config declares no languages')
     assert sorted(T._FAMILY_LANGUAGES['coqui']) == sorted(declared)
+
+
+# ── a language we advertise has to actually come out ─────────────────────────
+#
+# Advertising is only half the job. Probing all nine of Kokoro's languages
+# through the real endpoint on 2026-10-30 found eight working and ONE failing:
+# `fr` returned `espeak-ng has no such language`, because French is regioned in
+# espeak (fr-fr, fr-be, fr-ch) exactly as English is. The map's own comment
+# asserted the opposite -- that "it", "fr" and "de" all pass bare -- so the
+# claim had never been tested. de/it/nl/pl/tr/ru/ko/ar/cs/hu/sv do pass bare;
+# fr does not.
+
+def test_french_is_regioned_like_english():
+    """The regression: a bare 'fr' is what every OpenAI-compatible client
+    sends, and it 500ed."""
+    assert T.espeak_language("fr") == "fr-fr"
+
+
+@pytest.mark.parametrize("code", ["en", "fr", "pt", "zh"])
+def test_every_regioned_language_is_mapped(code):
+    """These four are the ones espeak refuses bare."""
+    assert T.espeak_language(code) != code
+
+
+@pytest.mark.parametrize("code", ["de", "it", "nl", "pl", "tr", "ru",
+                                  "ko", "ar", "cs", "hu", "sv"])
+def test_a_bare_code_espeak_accepts_is_left_alone(code):
+    """Measured: each of these works bare. Rewriting one would send espeak a
+    region it may not have."""
+    assert T.espeak_language(code) == code
+
+
+@pytest.mark.parametrize("code", ["fr-fr", "fr-be", "en-gb", "pt-br"])
+def test_an_explicit_region_passes_through(code):
+    assert T.espeak_language(code) == code
+
+
+def test_a_blank_language_defaults_to_english():
+    assert T.espeak_language("") == "en-us"
+    assert T.espeak_language(None) == "en-us"
+
+
+def test_case_and_underscores_are_normalised():
+    """Clients send en_US and EN-GB; both name a real espeak voice."""
+    assert T.espeak_language("FR") == "fr-fr"
+    assert T.espeak_language("en_GB") == "en-gb"
+
+
+def test_every_language_kokoro_advertises_resolves_to_something_espeak_takes():
+    """The listing and the synthesis path must agree. Each advertised code is
+    either accepted bare (measured) or mapped to a region that is.
+
+    This is the invariant that was broken: `fr` was advertised through the voice
+    ids and rejected at synthesis.
+    """
+    measured_bare_ok = {"es", "hi", "it", "ja", "en-gb", "en-us", "pt-br"}
+    for code in T.family_languages("hexgrad/Kokoro-82M"):
+        resolved = T.espeak_language(code)
+        assert resolved in measured_bare_ok or code in T._ESPEAK_REGIONLESS, (
+            f"Kokoro advertises {code!r} but it resolves to {resolved!r}, "
+            f"which was never measured against espeak")