Skip to content

armgit / meet

Public Imported from GitHub
0

Open source video conferencing app powered by LiveKit. Built with Django and React.

♻️(summary) refactor tasks signature and make transcription tz-aware

leoa695758da493
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 54a35e56..3fbec3ee 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -14,6 +14,10 @@ and this project adheres to
- ✨(helm) add support multiple transcribe worker / endpoint #1247
- ✨(backend) make LiveKit Egress recording encoding configurable #1288
+### Changed
+
+- ♻️(summary) change tasks endpoint signature
+
### Fixed
- ♻(frontend) standardize role terminology across localizations
diff --git a/src/backend/core/recording/event/notification.py b/src/backend/core/recording/event/notification.py
index 916c51d0..fee4922a 100644
--- a/src/backend/core/recording/event/notification.py
+++ b/src/backend/core/recording/event/notification.py
@@ -1,7 +1,9 @@
"""Service to notify external services when a new recording is ready."""
+import asyncio
import logging
import smtplib
+from datetime import datetime, timezone
from django.conf import settings
from django.core.mail import send_mail
@@ -10,8 +12,10 @@ from django.utils.translation import get_language, override
from django.utils.translation import gettext_lazy as _
import requests
+from asgiref.sync import async_to_sync
+from livekit import api as livekit_api
-from core import models
+from core import models, utils
logger = logging.getLogger(__name__)
@@ -131,7 +135,47 @@ class NotificationService:
return not has_failures
@staticmethod
- def _notify_summary_service(recording):
+ async def _get_recording_timestamps(worker_id):
+ """Fetch FileInfo.started_at and ended_at from LiveKit's egress API.
+
+ FileInfo.started_at is more accurate than EgressInfo.started_at because
+ it reflects when file recording actually began. The started_at value exposed
+ in the manifest file, as well as in the EgressInfo returned by the API,
+ corresponds to when the egress service received the request, not the moment
+ the egress worker effectively joined the room.
+
+ Returns:
+ Tuple of (started_at, ended_at) datetimes, either may be None.
+ """
+
+ if not worker_id:
+ return None, None
+
+ custom_configuration = {**settings.LIVEKIT_CONFIGURATION, "timeout": 10}
+ lkapi = utils.create_livekit_client(custom_configuration=custom_configuration)
+ try:
+ egress_list = await lkapi.egress.list_egress(
+ livekit_api.ListEgressRequest(egress_id=worker_id) # pylint: disable=no-member
+ )
+ except (livekit_api.TwirpError, OSError, asyncio.TimeoutError):
+ logger.exception("Could not fetch egress info for worker %s", worker_id)
+ return None, None
+ finally:
+ await lkapi.aclose()
+
+ if not egress_list.items or not egress_list.items[0].file_results:
+ logger.debug("No file_results for worker %s", worker_id)
+ return None, None
+
+ file_result = egress_list.items[0].file_results[0]
+
+ def _ns_to_utc(ns):
+ return datetime.fromtimestamp(ns / 1e9, tz=timezone.utc) if ns else None
+
+ return _ns_to_utc(file_result.started_at), _ns_to_utc(file_result.ended_at)
+
+ @staticmethod
+ def _notify_summary_service(recording: models.Recording):
"""Notify summary service about a new recording."""
if (
@@ -150,24 +194,37 @@ class NotificationService:
.first()
)
+ if settings.METADATA_COLLECTOR_ENABLED and recording.options.get(
+ "collect_metadata", False
+ ):
+ output_folder = settings.METADATA_COLLECTOR_OUTPUT_FOLDER
+ metadata_filename = f"{output_folder}/{recording.id}-metadata.json"
+ else:
+ metadata_filename = None
+
if not owner_access:
logger.error("No owner found for recording %s", recording.id)
return False
+
+ started_at, ended_at = async_to_sync(
+ NotificationService._get_recording_timestamps
+ )(recording.worker_id)
+
payload = {
"owner_id": str(owner_access.user.id),
- "filename": recording.key,
+ "recording_filename": recording.key,
+ "metadata_filename": metadata_filename, # For future use
"email": owner_access.user.email,
"sub": owner_access.user.sub,
"room": recording.room.name,
"language": recording.options.get("language"),
- "recording_date": recording.created_at.astimezone(
- owner_access.user.timezone
- ).strftime("%Y-%m-%d"),
- "recording_time": recording.created_at.astimezone(
- owner_access.user.timezone
- ).strftime("%H:%M"),
+ "owner_timezone": str(owner_access.user.timezone),
"download_link": f"{get_recording_download_base_url()}/{recording.id}",
"context_language": owner_access.user.language,
+ "recording_start_at": (started_at.isoformat() if started_at else None),
+ "recording_end_at": (
+ ended_at.isoformat() if ended_at else None
+ ), # For future use
}
headers = {
diff --git a/src/backend/meet/settings.py b/src/backend/meet/settings.py
index 5a594347..a855a20d 100755
--- a/src/backend/meet/settings.py
+++ b/src/backend/meet/settings.py
@@ -855,6 +855,11 @@ class Base(Configuration):
environ_name="METADATA_COLLECTOR_AGENT_NAME",
environ_prefix=None,
)
+ METADATA_COLLECTOR_OUTPUT_FOLDER = values.Value(
+ "metadata",
+ environ_name="METADATA_COLLECTOR_OUTPUT_FOLDER",
+ environ_prefix=None,
+ )
# External Applications
APPLICATION_ENABLED = values.BooleanValue(
diff --git a/src/summary/summary/api/route/tasks.py b/src/summary/summary/api/route/tasks.py
index 7645b4b5..91705f09 100644
--- a/src/summary/summary/api/route/tasks.py
+++ b/src/summary/summary/api/route/tasks.py
@@ -19,16 +19,18 @@ class TranscribeSummarizeTaskCreation(BaseModel):
"""Transcription and summarization parameters."""
owner_id: str
- filename: str
+ recording_filename: str
+ metadata_filename: Optional[str] = None
email: str
sub: str
version: Optional[int] = 2
room: Optional[str]
- recording_date: Optional[str]
- recording_time: Optional[str]
+ owner_timezone: Optional[str]
language: Optional[str]
download_link: Optional[str]
context_language: Optional[str] = None
+ recording_start_at: Optional[str] = None
+ recording_end_at: Optional[str] = None
@field_validator("language")
@classmethod
@@ -51,16 +53,18 @@ async def create_transcribe_summarize_task(request: TranscribeSummarizeTaskCreat
task = process_audio_transcribe_summarize_v2.apply_async(
args=[
request.owner_id,
- request.filename,
+ request.recording_filename,
+ request.metadata_filename,
request.email,
request.sub,
time.time(),
request.room,
- request.recording_date,
- request.recording_time,
+ request.owner_timezone,
request.language,
request.download_link,
request.context_language,
+ request.recording_start_at,
+ request.recording_end_at,
],
queue=settings.transcribe_queue,
)
diff --git a/src/summary/summary/core/analytics.py b/src/summary/summary/core/analytics.py
index d88b8e98..cffe1a37 100644
--- a/src/summary/summary/core/analytics.py
+++ b/src/summary/summary/core/analytics.py
@@ -112,7 +112,9 @@ class MetadataManager:
if self._is_disabled or self.has_task_id(task_id):
return
- _, filename, email, _, received_at, *_ = task_args
+ # Positional args mirror process_audio_transcribe_summarize_v2 signature:
+ # owner_id, recording_filename, metadata_filename, email, sub, received_at, ...
+ _, filename, _, email, _, received_at, *_ = task_args
start_time = time.time()
initial_metadata = {
diff --git a/src/summary/summary/core/celery_worker.py b/src/summary/summary/core/celery_worker.py
index fdd2c499..8b23fd1a 100644
--- a/src/summary/summary/core/celery_worker.py
+++ b/src/summary/summary/core/celery_worker.py
@@ -4,7 +4,6 @@
import json
import time
-from typing import Optional
import openai
import sentry_sdk
@@ -78,7 +77,7 @@ file_service = FileService()
def transcribe_audio(
*,
task_id: str,
- filename: str | None = None,
+ recording_filename: str | None = None,
language: str,
cloud_storage_url=None,
raises: bool = False,
@@ -90,7 +89,7 @@ def transcribe_audio(
Returns the transcription object, or None if the file could not be retrieved.
"""
- if bool(filename) == bool(cloud_storage_url):
+ if bool(recording_filename) == bool(cloud_storage_url):
raise ValueError(
"Either filename or cloud_storage_url must be provided, but not both."
)
@@ -105,11 +104,12 @@ def transcribe_audio(
# Transcription
try:
with file_service.prepare_audio_file(
- remote_object_key=filename,
+ remote_object_key=recording_filename,
cloud_storage_url=cloud_storage_url,
) as (audio_file, metadata):
metadata_manager.track(task_id, {"audio_length": metadata["duration"]})
+ # Compute language parameter
if language is None:
language = settings.whisperx_default_language
logger.info(
@@ -122,18 +122,21 @@ def transcribe_audio(
language,
)
+ # Call remote service for transcription
transcription_start_time = time.time()
-
transcription = whisperx_client.audio.transcriptions.create(
model=settings.whisperx_asr_model, file=audio_file, language=language
)
- transcription_time = round(time.time() - transcription_start_time, 2)
+ # Logging
+ transcription_duration = round(time.time() - transcription_start_time, 2)
metadata_manager.track(
task_id,
- {"transcription_time": transcription_time},
+ {"transcription_time": transcription_duration},
+ )
+ logger.info(
+ "Transcription received in %.2f seconds.", transcription_duration
)
- logger.info("Transcription received in %.2f seconds.", transcription_time)
logger.debug("Transcription: \n %s", transcription)
except FileServiceException as e:
@@ -148,7 +151,7 @@ def transcribe_audio(
"Unexpected error while preparing file | filename: %s "
"| cloud_storage_url: %s"
),
- filename,
+ recording_filename,
redacted_cloud_storage_url,
)
return None
@@ -162,8 +165,8 @@ def format_transcript(
context_language: str | None,
language: str,
room: str | None,
- recording_date: str | None,
- recording_time: str | None,
+ recording_datetime: str | None,
+ owner_timezone: str | None,
download_link: str | None,
) -> tuple[str, str]:
"""Format a transcription into readable content with a title.
@@ -179,8 +182,8 @@ def format_transcript(
return formatter.format(
transcription,
room=room,
- recording_date=recording_date,
- recording_time=recording_time,
+ recording_datetime=recording_datetime,
+ owner_timezone=owner_timezone,
download_link=download_link,
)
@@ -188,7 +191,7 @@ def format_transcript(
def format_actions(llm_output: dict) -> str:
"""Format the actions from the LLM output into a markdown list.
- fomat:
+ format:
- [ ] Action title Assignée à : assignee1, assignee2, Échéance : due_date
"""
lines = []
@@ -212,16 +215,18 @@ def format_actions(llm_output: dict) -> str:
def process_audio_transcribe_summarize_v2(
self,
owner_id: str,
- filename: str,
+ recording_filename: str,
+ metadata_filename: str | None,
email: str,
sub: str,
received_at: float,
- room: Optional[str],
- recording_date: Optional[str],
- recording_time: Optional[str],
- language: Optional[str],
- download_link: Optional[str],
- context_language: Optional[str] = None,
+ room: str | None,
+ owner_timezone: str | None,
+ language: str | None,
+ download_link: str | None,
+ context_language: str | None = None,
+ recording_start_at: str | None = None,
+ recording_end_at: str | None = None,
):
"""Process an audio file by transcribing it and generating a summary.
@@ -234,16 +239,20 @@ def process_audio_transcribe_summarize_v2(
Args:
self: Celery task instance (passed on with bind=True)
owner_id: Unique identifier of the recording owner.
- filename: Name of the audio file in MinIO storage.
+ recording_filename: Name of the audio file in MinIO storage.
+ metadata_filename: Name of the audio file in MinIO storage.
email: Email address of the recording owner.
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_date: Date of the recording (localized display string).
- recording_time: Time of the recording (localized display string).
+ 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.
context_language: ISO 639-1 language code of the meeting summary context text.
+ recording_start_at: ISO 8601 timestamp of when file recording actually started
+ (from LiveKit FileInfo.started_at via the egress_ended webhook).
+ recording_end_at: ISO 8601 timestamp of when file recording ended
+ (from LiveKit FileInfo.ended_at via the egress_ended webhook).
"""
logger.info(
"Notification received | Owner: %s | Room: %s",
@@ -253,19 +262,21 @@ def process_audio_transcribe_summarize_v2(
task_id = self.request.id
+ # Transcribe the audio
transcription = transcribe_audio(
- task_id=task_id, filename=filename, language=language
+ task_id=task_id, recording_filename=recording_filename, language=language
)
if transcription is None:
return
+ # Format output
content, title = format_transcript(
transcription,
context_language,
language,
room,
- recording_date,
- recording_time,
+ recording_start_at,
+ owner_timezone,
download_link,
)
diff --git a/src/summary/summary/core/transcript_formatter.py b/src/summary/summary/core/transcript_formatter.py
index 611806d2..4dec9e33 100644
--- a/src/summary/summary/core/transcript_formatter.py
+++ b/src/summary/summary/core/transcript_formatter.py
@@ -1,7 +1,9 @@
"""Transcript formatting into readable conversation format with speaker labels."""
import logging
-from typing import Optional, Tuple
+from datetime import datetime
+from typing import Tuple
+from zoneinfo import ZoneInfo
from summary.core.config import get_settings
from summary.core.locales import LocaleStrings
@@ -39,10 +41,10 @@ class TranscriptFormatter:
def format(
self,
transcription,
- room: Optional[str] = None,
- recording_date: Optional[str] = None,
- recording_time: Optional[str] = None,
- download_link: Optional[str] = None,
+ room: str | None = None,
+ recording_datetime: str | None = None,
+ owner_timezone: str | None = None,
+ download_link: str | None = None,
) -> Tuple[str, str]:
"""Format transcription into the final document and its title."""
segments = self._get_segments(transcription)
@@ -54,7 +56,7 @@ class TranscriptFormatter:
content = self._remove_hallucinations(content)
content = self._add_header(content, download_link)
- title = self._generate_title(room, recording_date, recording_time)
+ title = self._generate_title(room, recording_datetime, owner_timezone)
return content, title
@@ -83,7 +85,7 @@ class TranscriptFormatter:
return formatted_output
- def _add_header(self, content, download_link: Optional[str]) -> str:
+ def _add_header(self, content, download_link: str | None) -> str:
"""Add download link header to the document content."""
if not download_link:
return content
@@ -97,16 +99,20 @@ class TranscriptFormatter:
def _generate_title(
self,
- room: Optional[str] = None,
- recording_date: Optional[str] = None,
- recording_time: Optional[str] = None,
+ room: str | None = None,
+ recording_datetime: str | None = None,
+ owner_timezone: str | None = None,
) -> str:
"""Generate title from context or return default."""
- if not room or not recording_date or not recording_time:
+ if not room or not recording_datetime:
return self._locale.document_default_title
+ dt = datetime.fromisoformat(recording_datetime)
+ if owner_timezone:
+ dt = dt.astimezone(ZoneInfo(owner_timezone))
+
return self._locale.document_title_template.format(
room=room,
- room_recording_date=recording_date,
- room_recording_time=recording_time,
+ room_recording_date=dt.strftime("%Y-%m-%d"),
+ room_recording_time=dt.strftime("%H:%M"),
)
diff --git a/src/summary/tests/api/test_api_tasks.py b/src/summary/tests/api/test_api_tasks.py
index c32a96cb..07bffd1b 100644
--- a/src/summary/tests/api/test_api_tasks.py
+++ b/src/summary/tests/api/test_api_tasks.py
@@ -19,14 +19,14 @@ class TestTasks:
headers={"Authorization": "Bearer test-api-token"},
json={
"owner_id": "owner-123",
- "filename": "recording.mp4",
+ "recording_filename": "recording.mp4",
+ "metadata_filename": "metadata.json",
"email": "[email protected]",
"sub": "sub-123",
"room": "room-abc",
- "recording_date": "2026-01-01",
- "recording_time": "10:00:00",
+ "owner_timezone": "UTC",
"language": None,
- "download_link": "http://example.com/file.mp4",
+ "download_link": "https://example.com/file.mp4",
},
)
@@ -36,16 +36,18 @@ class TestTasks:
args = mock_apply_async.call_args.kwargs["args"]
assert args == [
"owner-123", # owner_id
- "recording.mp4", # filename
+ "recording.mp4", # recording_filename
+ "metadata.json", # metadata_filename
"[email protected]", # email
"sub-123", # sub
- 1735725600.0, # frozen time
+ 1735725600.0, # received_at
"room-abc", # room
- "2026-01-01", # recording_date
- "10:00:00", # recording_time
+ "UTC", # owner_timezone
None, # language
- "http://example.com/file.mp4", # download_link
+ "https://example.com/file.mp4", # download_link
None, # context_language
+ None, # recording_start_at
+ None, # recording_end_at
]
def test_create_task_invalid_language(self, client):