.

.env.example

@@ -33,7 +33,7 @@ OLLAMA_LLM_NUM_CTX=1024
 
 STT_PROVIDER=openai
 STT_FALLBACK_PROVIDER=elevenlabs
-# Empty means auto-detect the first answer, because the greeting asks for preferred language first.
+# Empty means auto-detect the first language-choice answer, then use Kazakh STT for name capture.
 STT_NAME_CAPTURE_LANGUAGE_CODE=
 STT_PROMPT=
 STT_PROMPT_RU=Это русская речь по телефону. Точно распознавай короткие ответы, имена, фамилии, слова благодарности и прощания. Если человек представился одним словом, сохраняй его как имя. Номера телефонов записывай цифрами.

core/session.py

@@ -395,6 +395,8 @@ _STT_ELEVENLABS_LANGUAGE_MAP = {
     "kk": "kaz",
 }
 
+DEFAULT_NAME_CAPTURE_STT_LANGUAGE_CODE = "kk"
+
 
 def _tts_language_for(session_language: str | None) -> str | None:
     if not session_language:
@@ -1789,17 +1791,21 @@ class CallSession:
         return raw or None
 
     async def _transcribe_batch_audio(self, audio_bytes: bytes) -> str:
-        if self._awaiting_customer_name and self._session_language is None and self._name_capture_language_code() is None:
+        if self._awaiting_customer_name:
+            configured_name_language = self._name_capture_language_code()
+            if self._session_language is None and configured_name_language is None:
                 return await self._transcribe_initial_language_choice(audio_bytes)
             return await self._stt.transcribe(
                 audio_bytes,
-            keyterms=self._name_capture_keyterms() if self._awaiting_customer_name else None,
-            language_code=(
-                self._name_capture_language_code()
-                if self._awaiting_customer_name
-                else _stt_language_for(self._session_language)
-            ),
-            force_batch=self._awaiting_customer_name,
+                keyterms=self._name_capture_keyterms(),
+                language_code=configured_name_language or DEFAULT_NAME_CAPTURE_STT_LANGUAGE_CODE,
+                force_batch=True,
+            )
+        return await self._stt.transcribe(
+            audio_bytes,
+            keyterms=None,
+            language_code=_stt_language_for(self._session_language),
+            force_batch=False,
         )
 
     async def _transcribe_initial_language_choice(self, audio_bytes: bytes) -> str: