diff --git a/app/providers/tts/piper.py b/app/providers/tts/piper.py
index f19cd04..75d2d59 100644
--- a/app/providers/tts/piper.py
+++ b/app/providers/tts/piper.py
@@ -21,6 +21,8 @@ from pathlib import Path
 
 from app.providers.tts.base import TTSProvider
 
+_GENDER_CODES = {"f", "m", "any"}
+
 try:  # bevorzugter Pfad: in-process, Modell bleibt geladen
     from piper import PiperVoice  # type: ignore
 
@@ -120,11 +122,23 @@ class PiperTTSProvider(TTSProvider):
         text: str,
         voice: str | None = None,
         audio_format: str = "pcm",
-        language: str | None = None,  # ignoriert: die Piper-Stimme kodiert die Sprache
+        language: str | None = None,
     ) -> bytes:
         if not text or not text.strip():
             raise ValueError("TTS input text is empty")
 
+        # Gender-Code (z. B. "f"/"m" aus Cartesia-Fallback) → echter Piper-Stimmname
+        if voice in _GENDER_CODES and language:
+            from app.dependencies import PIPER_VOICE_GENDERED, LANG_TO_PIPER_VOICE
+            lang = language.lower()
+            candidates = [voice, "f" if voice == "m" else "m"] if voice in ("f", "m") else ["f", "m"]
+            resolved = next(
+                (PIPER_VOICE_GENDERED.get((lang, g)) for g in candidates if PIPER_VOICE_GENDERED.get((lang, g))),
+                LANG_TO_PIPER_VOICE.get(lang),
+            )
+            if resolved:
+                voice = resolved
+
         voice_name, speaker_id = self._parse_speaker(voice)
         model, config = self._model_paths(voice_name)
         pcm, native_rate = await asyncio.to_thread(