feat: add Cantonese (粵語) language support via Whisper zh-yue

voxflow/app.py

@@ -479,7 +479,7 @@ def _build_quick_controls(self):
                      text_color=C["txt3"]).pack(anchor="w")
         self.lang_var = ctk.StringVar(value=self.config.language)
         ctk.CTkOptionMenu(
-            lf, values=["auto", "pl", "en", "de", "fr", "es", "it", "uk"],
+            lf, values=["auto", "pl", "en", "de", "fr", "es", "it", "uk", "zh-yue"],
             variable=self.lang_var,
             font=ctk.CTkFont(size=11),
             fg_color=C["bg_input"],
@@ -987,7 +987,7 @@ def _on_done(self, result: dict):
         _LANG_FLAGS = {
             "pl": "🇵🇱", "en": "🇬🇧", "de": "🇩🇪",
             "fr": "🇫🇷", "es": "🇪🇸", "it": "🇮🇹",
-            "uk": "🇺🇦",
+            "uk": "🇺🇦", "zh-yue": "🇭🇰",
         }
         flag = _LANG_FLAGS.get(lang, "🌍")
         dur = result.get("duration", 0)
@@ -1051,7 +1051,7 @@ def _refresh_history(self):
             _LANG_FLAGS = {
                 "pl": "🇵🇱", "en": "🇬🇧", "de": "🇩🇪",
                 "fr": "🇫🇷", "es": "🇪🇸", "it": "🇮🇹",
-                "uk": "🇺🇦",
+                "uk": "🇺🇦", "zh-yue": "🇭🇰",
             }
             flag = _LANG_FLAGS.get(e["language"], "🌍")
             preview = e["text"][:44] + ("…" if len(e["text"]) > 44 else "")

voxflow/config.py

@@ -18,7 +18,7 @@ def get_config_dir() -> Path:
 
 # Valid value ranges for security validation
 _VALID_MODELS = {"tiny", "base", "small", "medium", "large-v3"}
-_VALID_LANGUAGES = {"auto", "pl", "en", "de", "fr", "es", "it", "uk"}
+_VALID_LANGUAGES = {"auto", "pl", "en", "de", "fr", "es", "it", "uk", "zh-yue"}
 # Hotkey validation: allow any non-empty string (hotkey picker can produce
 # arbitrary key names). We only enforce it's a safe non-empty string.
 _VALID_TYPING_METHODS = {"clipboard", "keyboard"}
@@ -100,7 +100,7 @@ class VoxFlowConfig:
     """Application configuration."""
     # Model settings
     model_size: str = "small"
-    language: str = "auto"  # "auto", "pl", "en", "de", "fr", "es", "it", "uk"
+    language: str = "auto"  # "auto", "pl", "en", "de", "fr", "es", "it", "uk", "zh-yue"
     device: str = "cpu"  # "cpu" or "cuda"
     compute_type: str = "int8"  # "int8" for CPU, "float16" for GPU
 
@@ -181,6 +181,7 @@ def available_languages(self) -> dict:
             "es": "🇪🇸 Español",
             "it": "🇮🇹 Italiano",
             "uk": "🇺🇦 Українська",
+            "zh-yue": "🇭🇰 粵語 (Cantonese)",
         }
 
     @property

voxflow/post_processor.py

@@ -51,6 +51,11 @@
     "äh", "ähm", "hm", "öh", "öhm", "hmm", "ehm",
 }
 
+# Common Cantonese filler words/sounds that Whisper sometimes outputs
+CANTONESE_FILLERS = {
+    "嗯", "啊", "呀", "吖", "喎", "嘅", "喇", "咧", "咯",
+}
+
 # Common German Whisper corrections
 GERMAN_CORRECTIONS = {
     # Common split compounds
@@ -107,6 +112,8 @@ def post_process(
             result = _remove_fillers(result, POLISH_FILLERS)
         if language == "de":
             result = _remove_fillers(result, GERMAN_FILLERS)
+        if language == "zh-yue":
+            result = _remove_fillers(result, CANTONESE_FILLERS)
 
     # 4. Apply language-specific corrections
     if apply_corrections:
@@ -262,6 +269,13 @@ def _final_cleanup(text: str) -> str:
     "Polski tekst zawiera poprawne znaki diakrytyczne: ą, ć, ę, ł, ń, ó, ś, ź, ż."
 )
 
+CANTONESE_INITIAL_PROMPT = (
+    "Transcription of a recording in Cantonese (粵語). "
+    "The text is in traditional Cantonese Chinese. "
+    "Use traditional Chinese characters (繁體中文). "
+    "Common Cantonese words: 係, 唔, 幾, 咁, 喎, 呀, 吖, 喎, 嘅, 喇, 咧, 咯, 噶."
+)
+
 
 def get_initial_prompt(language: str) -> str:
     """Get the initial prompt for Whisper based on language.
@@ -275,5 +289,7 @@ def get_initial_prompt(language: str) -> str:
         return ENGLISH_INITIAL_PROMPT
     elif language == "de":
         return GERMAN_INITIAL_PROMPT
+    elif language == "zh-yue":
+        return CANTONESE_INITIAL_PROMPT
     else:
         return AUTO_INITIAL_PROMPT

voxflow/transcriber.py

@@ -190,7 +190,7 @@ def transcribe(
                 _LANG_FLAGS = {
                     "pl": "🇵🇱", "en": "🇬🇧", "de": "🇩🇪",
                     "fr": "🇫🇷", "es": "🇪🇸", "it": "🇮🇹",
-                    "uk": "🇺🇦",
+                    "uk": "🇺🇦", "zh-yue": "🇭🇰",
                 }
                 flag = _LANG_FLAGS.get(info.language, "🌍")
                 prob = info.language_probability * 100