fix(voice): keep asr partial polling off push path

services/ai_voice_runtime_service/media_runtime.py

@@ -291,9 +291,8 @@ class AudioSocketMediaRuntime:
         return normalized_intent != "unknown"
 
     def _should_emit_blind_ack(self, actor: MediaActor, pcm_bytes: bytes) -> bool:
-        if len(pcm_bytes) < self._immediate_ack_min_bytes:
-            return False
-        return actor.finalized_caller_turn_count > 0
+        del actor
+        return len(pcm_bytes) >= self._immediate_ack_min_bytes
 
     @staticmethod
     def _base_ack_variants(language: str | None, ack_kind: str) -> tuple[str, ...]:
@@ -325,8 +324,8 @@ class AudioSocketMediaRuntime:
             )
         return (
             "Угу, сейчас подскажу.",
-            "Мхм, я в контексте.",
-            "Ага, понял.",
+            "Мхм, секунду.",
+            "Ага, сейчас подскажу.",
             "Хм, секунду.",
             "Хорошо, сейчас подскажу.",
         )
@@ -485,8 +484,6 @@ class AudioSocketMediaRuntime:
                     stream_id,
                     bytes(batch),
                 )
-                with contextlib.suppress(StreamingASRUnavailable):
-                    await self._poll_streaming_partial(actor)
             except StreamingASRUnavailable as exc:
                 actor.asr_streaming_failed = True
                 actor.asr_streaming_enabled = False

tests/test_ai_voice_media_runtime.py

@@ -952,7 +952,7 @@ def test_media_runtime_voice_v2_emits_ack_for_short_utterance_after_reduced_thre
     assert speak_events[0][1] < timings["full_finished"]
 
 
-def test_media_runtime_voice_v2_skips_blind_ack_on_first_turn_without_partial_signal():
+def test_media_runtime_voice_v2_emits_blind_ack_on_first_turn_without_partial_signal():
     speak_events: list[tuple[str, float]] = []
 
     class _FastASRProvider(ASRProvider):
@@ -1042,8 +1042,9 @@ def test_media_runtime_voice_v2_skips_blind_ack_on_first_turn_without_partial_si
 
     asyncio.run(_scenario())
 
-    assert len(speak_events) == 1
-    assert speak_events[0][0] == "Подскажите подробнее, пожалуйста."
+    assert len(speak_events) == 2
+    assert speak_events[0][0] == runtime._ack_text("ru", "generic")
+    assert speak_events[1][0] == "Подскажите подробнее, пожалуйста."
 
 
 def test_media_runtime_ignores_low_signal_utterance_without_ack_or_turn():