Files
meetnote/api/app/services/transcribe.py
kang 3111c854c5 feat(api): A3 backend skeleton with FastAPI + SQLModel
- 3 tables: Meeting / TranscriptSegment / Summary (with state machine)
- Routes: /api/upload-url + /api/upload-complete + meetings CRUD
- MinIO presigned PUT for direct browser upload
- BackgroundTasks state-machine stub for A5 to flesh out
- SQLite for local dev, PostgreSQL+asyncpg for prod
- CORS configured for frontend on 4490

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-13 19:14:16 +08:00

47 lines
1.6 KiB
Python

"""Stub for the transcribe + summarize pipeline.
This module is intentionally a placeholder for A3. The real implementation
lands in A5 and will:
1. Download the object from MinIO
2. Probe duration and size
3. If file > 24 MB, ffmpeg silencedetect → split into <20 MB chunks
4. Call Groq Whisper for each chunk, shift timestamps, merge
5. Call Poe Claude with map-reduce for long audio
6. Persist segments + summary, update meeting.status to done
"""
import asyncio
from sqlmodel import select
from ..db import AsyncSessionLocal
from ..models import Meeting, MeetingStatus
async def process_meeting(meeting_id: int) -> None:
"""Background task launched after upload-complete.
For A3 we just walk the state machine so the frontend can see status
transitions; A5 swaps in real Groq/Poe calls.
"""
async with AsyncSessionLocal() as session:
meeting = await session.get(Meeting, meeting_id)
if not meeting:
return
try:
for status, delay in [
(MeetingStatus.splitting, 1),
(MeetingStatus.transcribing, 2),
(MeetingStatus.summarizing, 1),
(MeetingStatus.done, 0),
]:
meeting.status = status
if status == MeetingStatus.transcribing:
meeting.chunks_total = 1
meeting.chunks_done = 1
await session.commit()
await asyncio.sleep(delay)
except Exception as exc:
meeting.status = MeetingStatus.failed
meeting.error = str(exc)
await session.commit()