fix(omnivoice): use ISO language code instead of full name

containers/omnivoice/server.py

@@ -57,6 +57,7 @@ def health():
 def speech_clone(
     text: str = Form(...),
     language: str = Form(""),
+    language_code: str = Form(""),
     ref_text: str = Form(""),
     instruct: str = Form(""),
     speed: float = Form(1.0),
@@ -74,9 +75,10 @@ def speech_clone(
         tmp_path = tmp.name
 
     try:
+        # OmniVoice expects a 2-letter language code (en/it/...), not a full name
         audio = model.generate(
             text=text,
-            language=language or None,
+            language=language_code.strip().lower() or None,
             ref_audio=tmp_path,
             ref_text=ref_text or None,
             instruct=instruct or None,