wip: replace created_at by started_at. may want to revert

src/backend/core/recording/event/notification.py

@@ -166,7 +166,7 @@ class NotificationService:
             "sub": owner_access.user.sub,
             "room": recording.room.name,
             "language": recording.options.get("language"),
-            "recording_datetime": recording.created_at.isoformat(),
+            "worker_id": recording.worker_id,
             "owner_timezone": str(owner_access.user.timezone),
             "download_link": f"{get_recording_download_base_url()}/{recording.id}",
             "context_language": owner_access.user.language,

src/summary/summary/api/route/tasks.py

@@ -25,7 +25,7 @@ class TranscribeSummarizeTaskCreation(BaseModel):
     sub: str
     version: Optional[int] = 2
     room: Optional[str]
-    recording_datetime: Optional[str]
+    worker_id: Optional[str]
     owner_timezone: Optional[str]
     language: Optional[str]
     download_link: Optional[str]
@@ -58,7 +58,7 @@ async def create_transcribe_summarize_task(request: TranscribeSummarizeTaskCreat
             request.sub,
             time.time(),
             request.room,
-            request.recording_datetime,
+            request.worker_id,
             request.owner_timezone,
             request.language,
             request.download_link,

src/summary/summary/core/analytics.py

@@ -112,7 +112,7 @@ class MetadataManager:
         if self._is_disabled or self.has_task_id(task_id):
             return
 
-        _, filename, email, _, received_at, *_ = task_args
+        _, filename, _, email, _, received_at, *_ = task_args
 
         start_time = time.time()
         initial_metadata = {

src/summary/summary/core/celery_worker.py

@@ -3,8 +3,9 @@
 # ruff: noqa: PLR0913
 
 import json
+import os
 import time
-from datetime import datetime
+from datetime import datetime, timezone
 from typing import Optional
 
 import openai
@@ -182,7 +183,7 @@ def process_audio_transcribe_summarize_v2(
     sub: str,
     received_at: float,
     room: Optional[str],
-    recording_datetime: Optional[str],
+    worker_id: Optional[str],
     owner_timezone: Optional[str],
     language: Optional[str],
     download_link: Optional[str],
@@ -205,7 +206,7 @@ def process_audio_transcribe_summarize_v2(
         sub: OIDC subject identifier of the recording owner.
         received_at: Unix timestamp when the recording was received.
         room: room name where the recording took place.
-        recording_datetime: ISO 8601 UTC datetime of the recording.
+        worker_id: LiveKit egress ID used to fetch the egress JSON from S3.
         owner_timezone: IANA timezone of the recording owner (e.g. "Europe/Paris").
         language: ISO 639-1 language code for transcription.
         download_link: URL to download the original recording.
@@ -224,11 +225,28 @@ def process_audio_transcribe_summarize_v2(
     if transcription is None:
         return
 
+    # Fetch recording start time from LiveKit egress JSON
+    recording_datetime = None
+    recording_start_dt = None
+    if worker_id:
+        folder = os.path.dirname(recording_filename)
+        egress_json_key = f"{folder}/{worker_id}.json"
+        try:
+            egress_info = file_service.read_json(egress_json_key)
+            started_at_ns = egress_info.get("started_at")
+            if started_at_ns:
+                recording_start_dt = datetime.fromtimestamp(
+                    started_at_ns / 1e9, tz=timezone.utc
+                )
+                recording_datetime = recording_start_dt.isoformat()
+        except Exception:
+            logger.warning(
+                "Could not read egress JSON %s, falling back to no recording datetime",
+                egress_json_key,
+            )
+
     # Assign user names from metadata and transcription
     metadata = file_service.read_json(metadata_filename)
-    recording_start_dt = (
-        datetime.fromisoformat(recording_datetime) if recording_datetime else None
-    )
     if recording_start_dt is not None:
         assign_speakers(
             metadata, transcription, recording_start_dt, overlap_threshold=0.5

src/summary/summary/core/user_assign.py

@@ -64,7 +64,10 @@ class AssignmentResult:
 
         def _replace_speaker(item: dict[str, Any]) -> dict[str, Any]:
             if "speaker" in item and item["speaker"] in speaker_to_name:
-                return {**item, "speaker": speaker_to_name[item["speaker"]]}
+                return {
+                    **item,
+                    "speaker": f"{speaker_to_name[item['speaker']]}:{item['speaker']}",  # We still use the original diarization
+                }
             return item
 
         result: dict[str, Any] = {}
@@ -165,6 +168,7 @@ def _build_participant_timelines(
         ts = _parse_iso(event["timestamp"]).timestamp() - ref_epoch
         etype = event["type"]
 
+        # TODO: close open speech_start
         if etype == "speech_start":
             open_starts[pid] = max(ts, 0.0)
         elif etype == "speech_end":
@@ -224,6 +228,14 @@ def assign_speakers(
     )
     speaker_timelines = _build_speaker_timelines(transcription)
 
+    logger.info("Info info info")
+    logger.info(metadata)
+    logger.info(recording_start_datetime)
+    logger.info("participant_timelines")
+    logger.info(participant_timelines)
+    logger.info("participant_names")
+    logger.info(participant_names)
+
     result = AssignmentResult()
 
     for speaker, speaker_intervals in speaker_timelines.items():

src/summary/tests/api/test_api_tasks.py

@@ -24,7 +24,7 @@ class TestTasks:
                 "email": "user@example.com",
                 "sub": "sub-123",
                 "room": "room-abc",
-                "recording_datetime": "2026-01-01T10:00:00+00:00",
+                "worker_id": "EG_test123",
                 "owner_timezone": "UTC",
                 "language": None,
                 "download_link": "http://example.com/file.mp4",
@@ -43,7 +43,7 @@ class TestTasks:
             "sub-123",  # sub
             1735725600.0,  # frozen time
             "room-abc",  # room
-            "2026-01-01T10:00:00+00:00",  # recording_datetime
+            "EG_test123",  # worker_id
             "UTC",  # owner_timezone
             None,  # language
             "http://example.com/file.mp4",  # download_link