feat: self-hosted transcription + speaker diarization pipeline

[Copilot]

• Explore codebase and understand structure
• Add faster-whisper, pyannote.audio, ffmpeg-python dependencies to pyproject.toml
• Add transcription/diarization config env keys to app/core/config.py and env.example
• Create app/detection/transcribe.py with core pipeline functions
• Add TranscribeParams, TranscribeResultsModel and related sub-models
• Add transcribe to JobType enum and update CreateJobRequest param routing
• Add process_transcribe_task Celery task in app/core/tasks.py
• Register new task in app/core/celery_queue.py
• Add unit tests for merge/alignment logic (21 tests, all passing)
• Replace placeholder http://example.com/video.mp4 with real test video URL
• Add _transcribe_payload helper and transcribe job type tests to test_api.py
• Fix CI failures: test_status_returns_correct_shape and test_status_duplicate_project_returns_existing_job by adding a lightweight Redis job registry
• Regenerate openapi.json with all new transcribe/diarization schemas (TranscribeParams, TranscribeResultsModel, DiarizationSegmentModel, etc.)

---

💬 Send tasks to Copilot coding agent from Slack and Teams to turn conversations into code. Copilot posts an update in your thread when it's finished.

[younes200]

@copilot use this video url for testing instead of https://example.com/video.mp4 :

https://pt-storage.celluloid.cloud/web-videos/a53d2ce5-0db1-49bf-9daa-be2dafd46ee9-144.mp4

[Copilot]

@copilot use this video url for testing instead of https://example.com/video.mp4 :

https://pt-storage.celluloid.cloud/web-videos/a53d2ce5-0db1-49bf-9daa-be2dafd46ee9-144.mp4

Done in commit 7eed24a. Replaced the placeholder URL with https://pt-storage.celluloid.cloud/web-videos/a53d2ce5-0db1-49bf-9daa-be2dafd46ee9-144.mp4 in all three payload helpers (_object_detect_payload, _scene_detect_payload, _transcribe_payload) and the inline test bodies. Also added a _transcribe_payload helper and transcribe job type tests to test_api.py.

[younes200]

@copilot fix the failing github workflow action API Tests / api-tests (pull_request)

[Copilot]

@copilot fix the failing github workflow action API Tests / api-tests (pull_request)

Fixed in commit 68a1b35.

The two failing tests were caused by a race condition: with worker_prefetch_multiplier=1 and the real video URL (which takes ~60s to process), the single Celery worker is busy executing an earlier job. New jobs sit in the Celery queue in PENDING state and are invisible to inspect.active()/reserved(), so get_job_from_celery() returned None → 404.

The fix adds a lightweight Redis job registry (cvision:job:{job_id} hashes, 24h TTL) that stores minimal job metadata at enqueue time. When Celery reports PENDING and the inspect API finds nothing, we fall back to this registry. get_all_jobs() also merges registry entries so the deduplication check works correctly for all queued jobs.

[younes200]

@copilot generate and commit also the new openapi.json

[Copilot]

@copilot generate and commit also the new openapi.json

Done in commit 2a50283. The openapi.json is regenerated and now includes all new schemas: TranscribeParams, TranscribeResultsModel, DiarizationSegmentModel, TranscribeWordModel, and TranscriptSegmentModel, along with the transcribe job type in the CreateJobRequest discriminated union.