thomas.kopp 6b0ee60d94 fix: cache pyannote diarization pipeline to stop memory leak / OOM ...

The meeting pipeline created a fresh Diarizer per recording, each loading
the multi-GB pyannote speaker-diarization model anew (api/pipeline.py).
Whisper and Ollama run remotely in this deployment, so pyannote was the
only heavy in-process consumer. Reloading it per recording leaked CPU
memory (torch reference cycles + glibc arena fragmentation) that was
never returned to the OS, climbing to a 37 GB peak over a multi-day run
until the kernel OOM-killed the service.

Cache the loaded pipeline on the class and reuse it across Diarizer
instances, mirroring TranscriptionEngine._model. RSS now stays flat.

2026-07-22 09:03:42 +02:00

..

__init__.py

feat: config module with TOML defaults

2026-04-01 02:14:14 +02:00

test_alignment.py

feat: align_segments() — map Whisper timestamps to pyannote speakers

2026-04-02 01:00:58 +02:00

test_api.py

fix: update audio devices test to mock sounddevice instead of pactl

2026-04-02 07:52:34 +02:00

test_audio.py

feat: AudioRecorder accepts device param — reads audio.device from config

2026-04-01 20:32:44 +02:00

test_auth.py

feat: multi-user auth — per-user spaces, pbkdf2 passwords, session tokens, login page

2026-04-01 08:39:16 +02:00

test_config.py

feat: add diarization config defaults (enabled=false, hf_token)

2026-04-02 00:53:53 +02:00

test_diarization.py

fix: cache pyannote diarization pipeline to stop memory leak / OOM

2026-07-22 09:03:42 +02:00

test_llm.py

feat: OllamaClient.identify_speakers() and summarize() for diarization pipeline

2026-04-02 01:03:40 +02:00

test_output.py

feat: AI-generated title+tldr, subfolder structure, backlinks in transkript/zusammenfassung

2026-04-02 12:07:54 +02:00

test_transcription.py

feat: transcribe_file returns timestamped segments when with_segments=True

2026-04-02 00:55:53 +02:00