feat(1a): Flex-Sprachmodus vollständig entfernen

Sprache ist immer fest. Entfernt: language_mode aus Route/Schema/Config/
Runtime-Config, die Flex-Verzweigung in Orchestrator (effective_language =
language) und ws.py (immer normales transcribe), die "🔄 Flex"-Option im
Frontend samt zugehöriger app.js-Logik (langOverrides, loadMe, Turn-Sprache).

Die generische STT-Methode transcribe_detect bleibt als wiederverwendbare
Infrastruktur erhalten (nicht mehr aufgerufen). Test: Route trägt kein
language_mode mehr. 215 passed.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app/api/chat.py

@@ -50,7 +50,6 @@ async def chat(
         "input_endpoint": payload.input_endpoint,
         "output_endpoint": payload.output_endpoint,
         "language": payload.language,
-        "language_mode": payload.language_mode,
         "stt_provider": payload.stt_provider,
         "llm_provider": payload.llm_provider,
         "tts_provider": payload.tts_provider,
@@ -106,7 +105,6 @@ async def chat(
             voice=voice,
             output=output,
             history=llm_context,
-            language_mode=route.language_mode,
             text_only=bool(payload.text_only),
         )
     except Exception as exc:

app/api/ws.py

@@ -42,7 +42,6 @@ _OVERRIDE_KEYS = (
     "input_endpoint",
     "output_endpoint",
     "language",
-    "language_mode",
     "stt_provider",
     "llm_provider",
     "tts_provider",
@@ -67,7 +66,7 @@ async def _resolve(user, session_id, options):
 
 async def _run_turn(
     websocket, store, user, session_id, route, orchestrator, output, text, options,
-    detected_language: str | None = None, effective_voice: str | None = None,
+    effective_voice: str | None = None,
 ):
     """Faehrt einen Antwort-Turn und streamt die Events an den Client."""
     conversation = (
@@ -158,8 +157,6 @@ async def _run_turn(
                 history=llm_context,
                 on_token=on_token,
                 on_audio=on_audio,
-                language_mode=route.language_mode,
-                detected_language=detected_language,
                 text_only=text_only,
             )
         else:
@@ -169,8 +166,6 @@ async def _run_turn(
                 voice=voice,
                 output=output,
                 history=llm_context,
-                language_mode=route.language_mode,
-                detected_language=detected_language,
                 text_only=text_only,
             )
     except Exception as exc:
@@ -228,21 +223,9 @@ async def _voice_turn(websocket, store, user, session_id, audio, fmt, options):
         return
 
     try:
-        if route.language_mode == "flex":
-            # Flex: Sprache auto-erkennen, kein Übersetzen durch Whisper.
-            transcript, detected_language = await orchestrator.stt.transcribe_detect(
-                audio, fmt=fmt, language=None
-            )
-            # Stimme folgt der erkannten Sprache (Fallback: Routensprache).
-            effective_voice = voice_for_route(
-                route.tts_provider, detected_language or route.language, route.voice_gender
-            )
-        else:
-            # Fix: konfigurierte Sprache an Whisper → Whisper übersetzt automatisch.
-            transcript = await orchestrator.stt.transcribe(audio, fmt=fmt, language=route.language)
-            detected_language = None
-            # Stimme folgt der konfigurierten Sprache.
-            effective_voice = voice_for_route(route.tts_provider, route.language, route.voice_gender)
+        # Feste Sprache an Whisper → Whisper übersetzt automatisch.
+        transcript = await orchestrator.stt.transcribe(audio, fmt=fmt, language=route.language)
+        effective_voice = voice_for_route(route.tts_provider, route.language, route.voice_gender)
     except Exception as exc:
         await websocket.send_json({"type": "error", "status": 502, "detail": str(exc)})
         return
@@ -250,7 +233,6 @@ async def _voice_turn(websocket, store, user, session_id, audio, fmt, options):
     await websocket.send_json({
         "type": "transcript",
         "text": transcript,
-        "detected_language": detected_language,
     })
     if not transcript or not transcript.strip():
         await websocket.send_json({
@@ -260,7 +242,7 @@ async def _voice_turn(websocket, store, user, session_id, audio, fmt, options):
         return
     await _run_turn(
         websocket, store, user, session_id, route, orchestrator, output, transcript, options,
-        detected_language=detected_language, effective_voice=effective_voice,
+        effective_voice=effective_voice,
     )
 
 

app/config.py

@@ -112,7 +112,6 @@ class Settings(BaseSettings):
     openrouter_tts_voice: str = "alloy"
     openrouter_llm_model: str = "openai/gpt-4.1-mini"
     default_language: str = "de"
-    default_language_mode: str = "fix"
     default_input_endpoint: str = "local-default"
     default_output_endpoint: str = "local-default"
     default_stt_provider: str = "openrouter"

app/core/orchestrator.py

@@ -89,8 +89,6 @@ class Orchestrator:
         voice: str | None = None,
         output=None,
         history: list[dict] | None = None,
-        language_mode: str = "fix",
-        detected_language: str | None = None,
         text_only: bool = False,
     ):
         trace = PipelineTrace()
@@ -98,10 +96,8 @@ class Orchestrator:
         trace.raw_transcript = text
         trace.cleaned_transcript = await self.input_cleaner.run(text or "")
 
-        # Flex: die erkannte Sprache bestimmt die Ausgabe. Gibt es keine (z. B. Text-Chat
-        # ohne Audio), bleibt sie None -> das LLM antwortet von selbst in der Eingabesprache.
-        # Fix: immer die konfigurierte Sprache (Whisper hat die Eingabe übersetzt).
-        effective_language = detected_language if language_mode == "flex" else language
+        # Immer die konfigurierte (feste) Sprache (Whisper hat die Eingabe übersetzt).
+        effective_language = language
 
         with _stage("llm"):
             trace.semantic_response = await self.llm.complete(
@@ -144,8 +140,6 @@ class Orchestrator:
         history: list[dict] | None = None,
         on_token=None,
         on_audio=None,
-        language_mode: str = "fix",
-        detected_language: str | None = None,
         text_only: bool = False,
     ):
         """Wie chat_text, aber gestreamt.
@@ -158,9 +152,8 @@ class Orchestrator:
         trace.raw_transcript = text
         trace.cleaned_transcript = await self.input_cleaner.run(text or "")
 
-        # Siehe chat_text: Flex folgt der erkannten Sprache (None -> LLM spiegelt Eingabe),
-        # Fix nutzt die konfigurierte Sprache.
-        effective_language = detected_language if language_mode == "flex" else language
+        # Immer die konfigurierte (feste) Sprache.
+        effective_language = language
 
         # text_only: kein Server-Audio (Geräte-TTS spricht selbst) -> kein Chunked-TTS.
         chunker = SentenceChunker() if (on_audio and not text_only) else None

app/dependencies.py

@@ -155,7 +155,6 @@ ROUTE_KEYS = (
     "llm_provider",
     "tts_provider",
     "language",
-    "language_mode",
     "voice_gender",
 )
 
@@ -235,7 +234,6 @@ class ResolvedRoute:
     llm_provider: str
     tts_provider: str
     language: str
-    language_mode: str = "fix"
     voice_gender: str = "any"
 
     def as_dict(self) -> dict:
@@ -246,7 +244,6 @@ class ResolvedRoute:
             "llm_provider": self.llm_provider,
             "tts_provider": self.tts_provider,
             "language": self.language,
-            "language_mode": self.language_mode,
             "voice_gender": self.voice_gender,
         }
 
@@ -288,7 +285,6 @@ def resolve_route(
         "llm_provider": cfg.default_llm_provider,
         "tts_provider": cfg.default_tts_provider,
         "language": cfg.default_language,
-        "language_mode": cfg.default_language_mode,
     }
 
     user_prefs = user.prefs if user is not None else {}

app/runtime_config.py

@@ -20,7 +20,6 @@ RUNTIME_SETTABLE: dict[str, tuple[str, str, str]] = {
     "default_llm_provider":          ("LLM-Provider (Standard)",       "str",   "openrouter | local-openai-compatible"),
     "default_tts_provider":          ("TTS-Provider (Standard)",       "str",   "openrouter | piper | chatterbox — Standard, pro Nutzer überschreibbar"),
     "default_language":              ("Sprache (Standard)",            "str",   "de | en | … — Standard, pro Nutzer überschreibbar"),
-    "default_language_mode":         ("Sprachmodus (Standard)",        "str",   "fix | flex — Standard, pro Nutzer überschreibbar"),
     "openrouter_llm_model":          ("LLM-Modell (OpenRouter)",       "str",   "z.B. google/gemini-3.1-flash-lite"),
     "openrouter_tts_model":          ("TTS-Modell (OpenRouter)",       "str",   "z.B. google/gemini-3.1-flash-tts-preview"),
     "openrouter_tts_voice":          ("TTS-Stimme (OpenRouter)",       "str",   "z.B. Zephyr, Puck, Kore"),

app/schemas.py

@@ -46,7 +46,6 @@ class ChatRequest(BaseModel):
     input_endpoint: str | None = None
     output_endpoint: str | None = None
     language: str | None = None
-    language_mode: Literal["fix", "flex"] | None = None
     voice: str | None = None
     stt_provider: str | None = None
     llm_provider: str | None = None
@@ -93,7 +92,6 @@ class UserPrefs(BaseModel):
     llm_provider: str | None = None
     tts_provider: str | None = None
     language: str | None = None
-    language_mode: Literal["fix", "flex"] | None = None
 
 
 class MemoryCreate(BaseModel):

app/web/app.js

@@ -13,13 +13,9 @@ const cartesiaHint = $("#cartesia-lang-hint");
 let cartesiaUnsupportedLangs = [];   // von /api/config: Sprachen, die Cartesia nicht spricht
 const CARTESIA_LANG_LABELS = { ar: "Arabisch", it: "Italienisch", ru: "Russisch", pl: "Polnisch", nl: "Niederländisch", sv: "Schwedisch", tr: "Türkisch" };
 
-// "Flex" ist im Sprachmenü nur ein weiterer Wert: keine feste Sprache, sondern
-// automatische Erkennung. Eine konkrete Sprache = Fix-Modus (Eingabe wird übersetzt).
-// Beide internen Felder (language, language_mode) werden aus dieser einen Auswahl abgeleitet.
+// Die gewählte Sprache ist immer fest (Flex-Modus wurde entfernt).
 function langOverrides() {
-  const sel = langSel && langSel.value;
-  if (!sel || sel === "flex") return { language_mode: "flex" };
-  return { language: sel, language_mode: "fix" };
+  return { language: (langSel && langSel.value) || "de" };
 }
 
 // Nutzer-Präferenzen als Request-Overrides für jeden Turn.
@@ -348,10 +344,7 @@ function loadMe() {
       // Gespeicherte Nutzer-Präferenzen in die Dropdowns laden.
       const prefs = me.prefs || {};
       if (langSel) {
-        // Flex-Modus -> "flex", sonst die feste Sprache (Fallback: Flex).
-        langSel.value = prefs.language_mode === "flex"
-          ? "flex"
-          : (prefs.language || "de");
+        langSel.value = prefs.language || "de";
       }
       applyPlaybackChoice(prefs.tts_provider);
       applyGenderChoice(prefs.voice_gender || 'f');
@@ -544,10 +537,9 @@ function runTurn(path, onopen) {
     const pcm = [];
     let answerEl = null;
     let sampleRate = 24000;
-    // Sprache dieses Turns (für den Replay je Bubble): Flex folgt erkannter Sprache.
-    let routeLang = null, routeMode = null, detectedLang = null;
-    const turnLang = () =>
-      (routeMode === "flex" ? (detectedLang || routeLang) : routeLang) || null;
+    // Sprache dieses Turns (für den Replay je Bubble) = feste Routensprache.
+    let routeLang = null;
+    const turnLang = () => routeLang || null;
 
     ws.onopen = () => onopen(ws);
     ws.onerror = () => { statusEl.textContent = "Verbindungsfehler"; resolve(); };
@@ -560,10 +552,8 @@ function runTurn(path, onopen) {
       switch (msg.type) {
         case "ack":
           routeLang = (msg.route && msg.route.language) || null;
-          routeMode = (msg.route && msg.route.language_mode) || null;
           break;
         case "transcript":
-          detectedLang = msg.detected_language || null;
           if (msg.text) addMessage("user", msg.text, { lang: turnLang() });
           break;
         case "token":
@@ -620,9 +610,8 @@ async function sendText(text) {
   busy = true;
   setMicBusy();
   await ensureSession();
-  // Bei fester Sprache wird der eigene Text in dieser Sprache vorgelesen; bei Flex offen.
   const lo = langOverrides();
-  addMessage("user", text, { lang: lo.language_mode === "fix" ? lo.language : null });
+  addMessage("user", text, { lang: lo.language || null });
   statusEl.textContent = "denkt …";
   const audioStream = !isDeviceMode();  // Geräte-Modus: kein Server-Audio anfordern
   await runTurn("/ws/chat", (ws) => {

app/web/index.html

@@ -219,9 +219,8 @@
   <header class="shrink-0 h-14 flex items-center gap-2 px-3 border-b border-slate-200 dark:border-slate-700 bg-white/80 dark:bg-slate-800/80 backdrop-blur">
     <img src="/favicon.svg?v=1" alt="" class="w-7 h-7 shrink-0" />
     <h1 class="flex-1 truncate font-semibold text-base">Voice&nbsp;Assistant</h1>
-    <select id="lang-sel" title="Sprache — feste Sprache oder 🔄 Flex (folgt der gesprochenen Sprache)"
+    <select id="lang-sel" title="Sprache"
       class="text-sm rounded-lg border border-slate-300 dark:border-slate-600 bg-transparent px-2 py-1.5">
-      <option value="flex">🔄 Flex</option>
       <option value="de">🇩🇪 DE</option>
       <option value="en">🇬🇧 EN</option>
       <option value="fr">🇫🇷 FR</option>

tests/test_routing.py

@@ -132,3 +132,10 @@ def test_piper_falls_back_to_opposite_gender(monkeypatch):
     # (vor dem gender-agnostischen LANG_TO_PIPER_VOICE-Fallback).
     monkeypatch.delitem(PIPER_VOICE_GENDERED, ("de", "m"), raising=False)
     assert voice_for_route("piper", "de", "m") == "de_DE-kerstin-low"
+
+
+def test_route_has_no_language_mode():
+    # Flex-Modus wurde entfernt: die Route trägt kein language_mode mehr.
+    route = resolve_route()
+    assert "language_mode" not in route.as_dict()
+    assert not hasattr(route, "language_mode")