feat: add cache_fingerprint method to TTSProvider and its subclasses for voice/model configuration

services/ai_voice_runtime_service/ack_bank.py

@@ -49,6 +49,7 @@ class PrebakedAckBank:
     ) -> str:
         payload = {
             "provider": getattr(self._tts_provider, "name", "tts"),
+            "voice_fingerprint": self._tts_provider.cache_fingerprint(language, style_hints=style_hints),
             "language": str(language or "").strip() or None,
             "style_hints": style_hints or {},
             "text": text,

services/ai_voice_runtime_service/providers/tts.py

@@ -199,6 +199,16 @@ class TTSSynthesis:
 class TTSProvider:
     name = "stub"
 
+    def cache_fingerprint(self, language: str | None, *, style_hints: dict[str, object] | None = None) -> str:
+        """Identifies the exact voice/model config that would render `language`.
+
+        Callers that cache synthesized audio keyed only on (provider, language,
+        text) — like the prebaked ack bank — must fold this in too, or a voice
+        change silently keeps serving audio recorded with the old voice.
+        """
+        del language, style_hints
+        return ""
+
     def synthesize(
         self,
         text: str,
@@ -257,6 +267,10 @@ class OpenAITTSProvider(TTSProvider):
     def _voice(self, language: str | None) -> str:
         return self._kz_voice if _normalize_voice_language(language) == "kz" else self._ru_voice
 
+    def cache_fingerprint(self, language: str | None, *, style_hints: dict[str, object] | None = None) -> str:
+        del style_hints
+        return f"{self._voice(language)}:{self._model(language)}:{self._speed}"
+
     def _cache_key(self, text: str, *, language: str | None) -> str:
         payload = {
             "provider": self.name,
@@ -433,6 +447,12 @@ class YandexTTSProvider(TTSProvider):
             return explicit_role
         return self._role
 
+    def cache_fingerprint(self, language: str | None, *, style_hints: dict[str, object] | None = None) -> str:
+        return (
+            f"{self._voice(language)}:{self._effective_role(style_hints=style_hints)}:"
+            f"{self._speed}:{self._sample_rate_hz}"
+        )
+
     def _cache_key(self, text: str, *, language: str | None, style_hints: dict[str, object] | None = None) -> str:
         payload = {
             "provider": self.name,
@@ -643,6 +663,13 @@ class ElevenLabsTTSProvider(TTSProvider):
     def _language_code(self, language: str | None) -> str:
         return self._kz_language_code if _normalize_voice_language(language) == "kz" else self._ru_language_code
 
+    def cache_fingerprint(self, language: str | None, *, style_hints: dict[str, object] | None = None) -> str:
+        del style_hints
+        return (
+            f"{self._voice(language)}:{self._model(language)}:"
+            f"{self._language_code(language)}:{self._output_format}"
+        )
+
     def _cache_key(
         self,
         text: str,