.

core/session.py

@@ -557,6 +557,13 @@ def _sanitize_voice_text(text: str) -> str:
 
 DEFAULT_NAME_RETRY_TEXT = "Подскажите, пожалуйста, как я могу к вам обращаться?"
 DEFAULT_PERSONALIZED_GREETING_TEMPLATE = '{name}, я работаю на основе искусственного интеллекта и постараюсь максимально внимательно разобраться с вашим вопросом. Чем я могу помочь?'
+_NAME_CAPTURE_CONFUSION_ALIASES = {
+    "парень": "Арнур",
+    "арнер": "Арнур",
+    "ар нур": "Арнур",
+    "шомарт": "Жомарт",
+    "шумарт": "Жомарт",
+}
 
 
 def _voice_text_key(text: str | None) -> str:
@@ -564,6 +571,9 @@ def _voice_text_key(text: str | None) -> str:
     return re.sub(r"\s+", " ", compact).strip()
 
 
+_KAZAKH_NAME_BY_KEY = {_voice_text_key(name): name for name in TOP_KAZAKH_NAMES}
+
+
 def _voice_letters_key(text: str | None) -> str:
     return "".join(re.findall(r"[^\W\d_]+", str(text or "").lower(), flags=re.UNICODE))
 
@@ -582,6 +592,39 @@ def _looks_like_supported_name_text(text: str) -> bool:
     )
 
 
+def _correct_name_capture_candidate(name: str | None) -> str | None:
+    candidate = _canonical_name(str(name or "").strip())
+    if not candidate:
+        return None
+    key = _voice_text_key(candidate)
+    exact_match = _KAZAKH_NAME_BY_KEY.get(key)
+    if exact_match:
+        return exact_match
+    alias_match = _NAME_CAPTURE_CONFUSION_ALIASES.get(key)
+    if alias_match:
+        return alias_match
+
+    best_name = None
+    best_ratio = 0.0
+    for gazetteer_key, gazetteer_name in _KAZAKH_NAME_BY_KEY.items():
+        ratio = difflib.SequenceMatcher(None, key, gazetteer_key).ratio()
+        if ratio > best_ratio:
+            best_ratio = ratio
+            best_name = gazetteer_name
+    if best_name is None:
+        return candidate
+    threshold = 0.82 if len(key) >= 5 else 0.9
+    if best_ratio >= threshold:
+        LOGGER.info(
+            "name capture candidate corrected via Kazakh names: candidate=%r corrected=%r ratio=%.3f",
+            candidate,
+            best_name,
+            best_ratio,
+        )
+        return best_name
+    return candidate
+
+
 def _normalize_name_candidate(text: str | None) -> str | None:
     raw = str(text or "").strip(" \t\r\n,.;:!?\"'()[]{}")
     if not raw or any(ch.isdigit() for ch in raw):
@@ -798,6 +841,16 @@ def _voice_start_name_outcome(text: str | None) -> tuple[str, str | None]:
     if not raw:
         return "name_not_obtained", None
     candidate, needs_followup = _extract_name_candidate(raw)
+    if candidate:
+        corrected_candidate = _correct_name_capture_candidate(candidate) or candidate
+        if corrected_candidate != candidate:
+            LOGGER.info(
+                "name capture candidate normalized: candidate=%r corrected=%r transcript=%r",
+                candidate,
+                corrected_candidate,
+                _preview_text(raw),
+            )
+        candidate = corrected_candidate
     if candidate and not needs_followup:
         return "name_obtained", candidate
     if candidate:
@@ -1493,6 +1546,8 @@ class CallSession:
                     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 None,
+                        force_batch=self._awaiting_customer_name,
                     )
                 ).strip()
             except Exception:
@@ -1513,6 +1568,14 @@ class CallSession:
     def _name_capture_keyterms(self) -> tuple[str, ...]:
         return TOP_KAZAKH_NAMES
 
+    def _name_capture_language_code(self) -> str | None:
+        raw = (
+            os.getenv("STT_NAME_CAPTURE_LANGUAGE_CODE", "").strip()
+            or os.getenv("STT_NAME_CAPTURE_LANGUAGE", "").strip()
+            or "kz"
+        )
+        return raw or None
+
     def _build_name_collection_response(self, transcript: str) -> str | None:
         if not self._awaiting_customer_name:
             return None
@@ -2074,6 +2137,8 @@ class CallSession:
         return await self._stt.transcribe(
             utterance_audio,
             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 None,
+            force_batch=self._awaiting_customer_name,
         )
 
     def _maybe_dump_audio(self, audio_bytes: bytes) -> None: