회기 무발화 0턴 분리, 자기예측 락 불변식 및 TDD 회귀 검증 완료

apps/api/app/services/voice.py

@@ -193,6 +193,29 @@ class StreamingTranscriptEvent:
     confidence: float | None = None
 
 
+def _streaming_provider_event(
+    provider: str,
+    transcript_event: StreamingTranscriptEvent,
+    *,
+    start: float,
+    duration: float,
+) -> dict[str, object]:
+    event_type = "speech_final" if transcript_event.speech_final else (
+        "speech_end" if transcript_event.final else "voice_activity"
+    )
+    provider_event: dict[str, object] = {
+        "type": event_type,
+        "provider": provider,
+        "source": "streaming_stt",
+        "start_ms": round(start * 1000),
+        "duration_ms": round(duration * 1000),
+        "is_final": transcript_event.final,
+    }
+    if transcript_event.confidence is not None:
+        provider_event["confidence"] = transcript_event.confidence
+    return provider_event
+
+
 class DeepgramStreamingSession:
     """One Deepgram Listen WebSocket, scoped to exactly one learner utterance."""
 
@@ -364,29 +387,21 @@ class DeepgramStreamingSession:
         if not display_text and not speech_final:
             return
 
-        event_type = "speech_final" if speech_final else (
-            "speech_end" if is_final else "voice_activity"
+        transcript_event = StreamingTranscriptEvent(
+            text=display_text,
+            final=is_final,
+            speech_final=speech_final,
+            confidence=confidence,
+        )
+        provider_event = _streaming_provider_event(
+            "deepgram",
+            transcript_event,
+            start=start,
+            duration=duration,
         )
-        provider_event: dict[str, object] = {
-            "type": event_type,
-            "provider": "deepgram",
-            "source": "streaming_stt",
-            "start_ms": round(start * 1000),
-            "duration_ms": round(duration * 1000),
-            "is_final": is_final,
-        }
-        if confidence is not None:
-            provider_event["confidence"] = confidence
         if is_final or speech_final:
             self._provider_events.append(provider_event)
-        await self._on_event(
-            StreamingTranscriptEvent(
-                text=display_text,
-                final=is_final,
-                speech_final=speech_final,
-                confidence=confidence,
-            )
-        )
+        await self._on_event(transcript_event)
 
     def _consume_final_words(self, value: object) -> None:
         if not isinstance(value, list):
@@ -620,29 +635,21 @@ class LocalWhisperStreamingSession:
         if not display_text and not speech_final:
             return
 
-        event_type = "speech_final" if speech_final else (
-            "speech_end" if is_final else "voice_activity"
+        transcript_event = StreamingTranscriptEvent(
+            text=display_text,
+            final=is_final,
+            speech_final=speech_final,
+            confidence=confidence,
+        )
+        provider_event = _streaming_provider_event(
+            "local_whisper",
+            transcript_event,
+            start=start,
+            duration=duration,
         )
-        provider_event: dict[str, object] = {
-            "type": event_type,
-            "provider": "local_whisper",
-            "source": "streaming_stt",
-            "start_ms": round(start * 1000),
-            "duration_ms": round(duration * 1000),
-            "is_final": is_final,
-        }
-        if confidence is not None:
-            provider_event["confidence"] = confidence
         if is_final or speech_final:
             self._provider_events.append(provider_event)
-        await self._on_event(
-            StreamingTranscriptEvent(
-                text=display_text,
-                final=is_final,
-                speech_final=speech_final,
-                confidence=confidence,
-            )
-        )
+        await self._on_event(transcript_event)
 
     def _consume_final_words(self, value: object, *, offset: float) -> None:
         if not isinstance(value, list):