Meeting Record - Private Voice Recording & Transcription

Local voice recording with live Whisper transcription. All processing happens on your machine - no cloud APIs, no data leaves your computer.

Why This Exists

Most "AI voice recorders" (Plaud Note knockoffs, etc.) are just microphones that send your audio to cloud servers for processing. You're paying for hardware that:

Sends your private conversations to unknown servers
Requires monthly subscriptions after free tier
May store/analyze your data for training or other purposes
Stops working if the company shuts down

Meeting Record does the same thing, but locally. Your audio never leaves your machine.

Features

Continuous recording - Full audio backup of entire meeting
Live transcription - See transcripts as you speak (10-second chunks)
Final transcription - High-quality full transcription when you stop
100% local - Uses whisper.cpp, no internet required
No subscriptions - Free forever, you own it

Installation

./install.sh

The installer will:

Install system dependencies (ffmpeg, pulseaudio-utils, build tools)
Clone and build whisper.cpp
Download the Whisper model (base.en by default - fast & good)
Install the meeting-record scripts to ~/bin/

Custom Options

# Use different model (tiny, base, small, medium, large)
WHISPER_MODEL=small.en ./install.sh

# Custom install location
WHISPER_CPP_DIR=/opt/whisper.cpp ./install.sh

# Custom bin directory
BIN_DIR=/usr/local/bin ./install.sh

Usage

# Start recording (auto-named with timestamp)
meeting-record

# Named recording
meeting-record "client-call-acme-corp"

# 30-second test to verify setup
meeting-record-test

Press Ctrl+C to stop recording. The script will:

Stop audio capture
Generate a high-quality final transcript
Show you where all files are saved

Output Files

After recording a meeting named "client-call", you'll find:

~/Recordings/meetings/client-call/
├── full-audio.wav              # Complete recording (your backup)
├── live-transcript.txt         # Real-time chunked transcription
├── final-transcript.txt        # High-quality full transcription
├── chunks/
│   ├── chunk-0000.wav
│   ├── chunk-0001.wav
│   └── ...
└── transcripts/
    ├── chunk-0000.txt
    ├── chunk-0001.txt
    └── ...

Requirements

Linux (Debian/Ubuntu/Fedora/Arch)
PulseAudio or PipeWire (for audio capture)
~500MB disk space for whisper.cpp + model

Model Options

Model	Size	Speed	Quality	Best For
tiny.en	75MB	Fastest	Basic	Quick notes
base.en	148MB	Fast	Good	Recommended
small.en	488MB	Medium	Better	Important meetings
medium.en	1.5GB	Slow	Great	Accuracy-critical
large	3GB	Slowest	Best	Maximum accuracy

Change model with: WHISPER_MODEL=small.en ./install.sh

Troubleshooting

"parecord not found"

sudo apt install pulseaudio-utils  # Debian/Ubuntu
sudo dnf install pulseaudio-utils  # Fedora

"No audio being captured"

Check your default audio source: pactl list sources
Make sure microphone is not muted

"whisper-cli not found"

The installer should build this. If it failed, check cmake output
You can rebuild: cd ~/opt/whisper.cpp/build && cmake --build .

"meeting-record: command not found"

Add ~/bin to PATH: export PATH="$HOME/bin:$PATH"
Add this to your ~/.bashrc or ~/.zshrc

Privacy Comparison

Feature	Cloud AI Recorders	Meeting Record
Audio processing	Remote servers	Local only
Data storage	Their servers	Your disk
Internet required	Yes	No
Monthly fees	Usually	Never
Works offline	No	Yes
You control data	No	Yes

License

MIT - Do whatever you want with it.

Part of the Privacy Toolkit - Tools for digital sovereignty.

4 KiB Raw Permalink Blame History