Persist realtime transcription sessions to history

The websocket handler now accumulates the full session's PCM audio
(not just the per-chunk buffer that gets discarded after each
worker call) and, on stop, writes it as a wav to UPLOAD_DIR plus a
result JSON to RESULT_DIR via the new _save_realtime_history()
helper — same shape /transcribe already writes, tagged
gateway_backend="realtime" so the history table visually
distinguishes it from file uploads. This reuses /asr/history and
/asr/uploads/{filename} as-is; no new endpoints needed.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

app/asr/router.py

@@ -268,6 +268,41 @@ async def _send_chunk(wav_bytes: bytes, *, model: str, language: str, beam_size:
     return resp.json()
 
 
+def _save_realtime_history(
+    *,
+    all_audio: bytearray,
+    text: str,
+    segments: List[Dict[str, Any]],
+    model: str,
+    language: str,
+    duration: float,
+) -> Optional[str]:
+    if not all_audio:
+        return None
+
+    ensure_runtime_dirs()
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S_%f")
+    saved_upload = UPLOAD_DIR / f"{timestamp}.wav"
+    saved_upload.write_bytes(_pcm_float32_to_wav(bytes(all_audio)))
+
+    payload = {
+        "backend": "faster-whisper",
+        "gateway_backend": "realtime",
+        "model": model,
+        "language": language,
+        "duration": round(duration, 3),
+        "text": text,
+        "segments": segments,
+        "diarized": False,
+        "_filename": "실시간 녹음",
+        "_timestamp": timestamp,
+        "upload_file": str(saved_upload),
+    }
+    result_path = RESULT_DIR / f"{timestamp}.json"
+    result_path.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
+    return timestamp
+
+
 @ws_router.websocket("/ws/realtime")
 async def realtime_ws(websocket: WebSocket) -> None:
     if not websocket.session.get("user"):
@@ -276,6 +311,7 @@ async def realtime_ws(websocket: WebSocket) -> None:
     await websocket.accept()
 
     audio_buf = bytearray()
+    all_audio = bytearray()
     partial_texts: List[str] = []
     all_segments: List[Dict[str, Any]] = []
     time_offset = 0.0
@@ -333,6 +369,7 @@ async def realtime_ws(websocket: WebSocket) -> None:
                 break
             if "bytes" in data:
                 audio_buf.extend(data["bytes"])
+                all_audio.extend(data["bytes"])
                 while len(audio_buf) >= chunk_bytes:
                     await process_buffer(bytearray(audio_buf[:chunk_bytes]))
                     audio_buf = audio_buf[chunk_bytes:]
@@ -341,11 +378,21 @@ async def realtime_ws(websocket: WebSocket) -> None:
                 if msg.get("cmd") == "stop":
                     if audio_buf:
                         await process_buffer(audio_buf)
+                    full_text = " ".join(x for x in partial_texts if x)
+                    result_id = _save_realtime_history(
+                        all_audio=all_audio,
+                        text=full_text,
+                        segments=all_segments,
+                        model=model,
+                        language=language,
+                        duration=time_offset,
+                    )
                     await websocket.send_json({
                         "type": "final",
-                        "text": " ".join(x for x in partial_texts if x),
+                        "text": full_text,
                         "segments": all_segments,
                         "diarized": False,
+                        "id": result_id,
                     })
                     break
 

app/ui/assets/js/asr.js

@@ -200,7 +200,7 @@ async function startRecording() {
       appendLive('');
     } else if (msg.type === 'final') {
       rtFinalText.value = msg.text || '';
-      setIndicator('idle', '완료');
+      setIndicator('idle', msg.id ? '완료 (처리 내역에 저장됨)' : '완료');
       cleanupAudio();
     } else if (msg.type === 'error') {
       setIndicator('idle', `오류: ${msg.detail}`);