I don't listen to meetings anymore. My sloth does.

Description and use

Durmain is a local assistant that records call audio, transcribes with faster‑whisper, detects the user’s name, and after the call ends produces minutes using Ollama’s Qwen2.5 7B, keeping all data on the laptop.

Problem

Taking notes in a meeting is expensive and often done poorly because the speaker is simultaneously being transcribed and listened to, and most transcription tools send audio to third‑party servers.

Outcome

Durmain achieved 8/10 name detections with 0 false alarms in live calls, 16/20 in scripted recordings, produced minutes in ~36 s for a 20‑minute lecture, and used 583 MB RAM with CPU spikes of 40‑50%.

Workflow

• Record system audio via pyaudiowpatch in 5‑second blocks
• Transcribe blocks with faster‑whisper (model small, int8) and VAD
• Detect user’s name in transcript using rapidfuzz
• When name detected, display sloth overlay and start live note taking
• After call ends, send transcript to Ollama (qwen2.5:7b) to produce minutes
• Save minutes locally

Evidence

Durmain listens to your computer's audio, transcribes it locally, and stays asleep until your name comes up.

Source

The pipeline is simple: system audio ─► 5 s blocks ─► faster-whisper + VAD ─► transcript │ │ ▼ ▼ when the call ends name detector ─► sloth Ollama (qwen2.5:7b) ─► minutes

Source

Everything runs on your machine. No audio or text ever leaves it.

Source

The first time I paused a video, the transcript said "Legendas pela comunidade Amara.org"… Whisper learned from subtitled videos, so silence turns into subtitle credits.

Source

I built this with an AI coding assistant acting as a mentor… wrote some of the glue code (the Ollama call, the overlay) so I could make the deadline.

Source