Hearsay

Windows desktop app that records system audio and/or microphone input and transcribes it in real-time using OpenAI's open-source Whisper model running locally.

No API calls, no cloud services -- everything runs on your machine.

---

Features

• System audio capture -- record what your speakers play (YouTube, Teams, podcasts, etc.) via WASAPI loopback
• Microphone capture -- record from your mic, or capture both sources together
• Speaker-source labels -- transcripts mark every switch between Remote (system audio) and Local (microphone) speech with a bold label, so conversations read like a dialogue
• Real-time transcription -- text appears in a live view window as you record
• Local AI -- uses faster-whisper (CTranslate2), no internet required after model download
• GPU + CPU -- auto-detects NVIDIA GPU; works on CPU with INT8 quantization
• Markdown output -- date-stamped .md transcripts saved to your chosen directory
• System tray app -- runs quietly in the tray, right-click to start/stop recording
• First-run wizard -- detects hardware, downloads the right model, configures everything
• Windows installer -- appears in Add/Remove Programs, Start Menu shortcut, clean uninstall

Quick Start

From source

# Clone the repo
git clone https://github.com/parkscloud/Hearsay.git
cd Hearsay

# Install dependencies
pip install -r requirements.txt

# Run
python -m hearsay

On first launch, the setup wizard walks you through:

1. Hardware detection (GPU vs CPU)
2. Audio source selection
3. Output directory
4. Model download

After setup, Hearsay lives in your system tray. Right-click the icon to start recording.

Installed version

Download the latest installer from the Releases page and run HearsaySetup.exe. The app appears in your Start Menu and Add/Remove Programs.

Silent install (RMM / SCCM / Intune)

HearsaySetup.exe /VERYSILENT /SUPPRESSMSGBOXES /NORESTART

Installs to C:\Program Files\Hearsay for all users. Hearsay starts automatically at login. To skip auto-start:

HearsaySetup.exe /VERYSILENT /SUPPRESSMSGBOXES /NORESTART /TASKS=""

Uninstall silently:

"C:\Program Files\Hearsay\unins000.exe" /VERYSILENT

Usage

1. Right-click the tray icon
2. Choose Start Recording > System Audio, Microphone, or Both
3. Audio is transcribed in real-time -- open Live Transcript to watch
4. Stop Recording when done -- a timestamped .md file is saved to your output directory

In Both mode each source is transcribed separately and labeled (Remote for system audio, Local for your mic). Headphones give the cleanest separation; on speakers, Hearsay automatically filters the microphone's echo of remote speech.

Hardware Requirements

Setup	Recommended Model	Speed
NVIDIA GPU (6+ GB VRAM)	turbo (float16)	~8x real-time
NVIDIA GPU (4 GB VRAM)	small.en (float16)	~4x real-time
CPU only	small.en (int8)	~1x real-time

A 1-hour recording transcribes in ~7 min on GPU or ~60 min on CPU.

Project Structure

src/hearsay/
├── __init__.py              # Version string
├── __main__.py              # Entry point
├── app.py                   # Application orchestrator
├── config.py                # AppConfig + ConfigManager (JSON in %APPDATA%)
├── constants.py             # App name, model table, defaults
├── audio/
│   ├── devices.py           # Enumerate loopback + mic devices
│   ├── recorder.py          # AudioRecorder thread
│   └── resampler.py         # Resample to 16kHz mono float32
├── transcription/
│   ├── gpu_detect.py        # Detect CUDA, recommend model
│   ├── model_manager.py     # Download and cache Whisper models
│   ├── engine.py            # TranscriptionEngine (faster-whisper)
│   └── pipeline.py          # TranscriptionPipeline thread
├── output/
│   ├── formatter.py         # Timestamp formatting
│   └── markdown_writer.py   # Write .md transcripts
├── ui/
│   ├── tray.py              # System tray icon (pystray)
│   ├── wizard.py            # First-run setup wizard
│   ├── live_view.py         # Live transcript window
│   ├── about_window.py      # About dialog
│   ├── settings_window.py   # Settings editor
│   ├── window_icon.py       # Title-bar icon helper (Hearsay logo)
│   ├── icons.py             # Programmatic icon generation
│   └── theme.py             # customtkinter theme
└── utils/
    ├── paths.py             # %APPDATA%\Hearsay directories
    ├── logging_setup.py     # File + console logging
    └── threading_utils.py   # StoppableThread, safe_after

Building

Prerequisites

1. Python 3.11+ (64-bit) -- the installer packages x64-only builds
2. Project dependencies: pip install -r requirements.txt
3. PyInstaller: pip install pyinstaller
4. Inno Setup 6+: winget install JRSoftware.InnoSetup

Build steps

# 1. Bundle the app with PyInstaller (output in dist\Hearsay\)
build.bat

# 2. Compile the Windows installer (winget does not add ISCC to PATH)
"C:\Program Files (x86)\Inno Setup 6\ISCC.exe" installer.iss

The installer is written to installer_output\HearsaySetup.exe.

The PyInstaller bundle is defined by Hearsay.spec; build.bat is a thin wrapper around it. Edit the spec to change bundled data, hidden imports, or the icon.

Releasing

See RELEASING.md for instructions on creating GitHub releases with the installer attached. This is a reference for the author and LLM when updating the application.

Tech Stack

• Python 3.11+
• faster-whisper -- CTranslate2-based Whisper inference
• PyAudioWPatch -- WASAPI loopback recording
• sounddevice -- Microphone capture
• customtkinter -- Modern UI
• pystray + Pillow -- System tray
• PyInstaller + Inno Setup -- Build and install

Contact

Robert Parks
raparks.com

License

MIT -- free to use, modify, and distribute for any purpose (personal or commercial). See LICENSE for full text.