Moon Live Captions

日本語版はこちら

Moon Live Captions is a Windows desktop application that shows real-time speech-to-text captions on screen. Audio is captured from the system speaker output and/or microphone and transcribed locally using the on-device Moonshine neural network — no internet connection or cloud service is required during operation.

Features

• Real-time transcription of system audio (speaker) and/or microphone
• Multiple languages — Japanese, English, Spanish, Korean, Chinese, Arabic, Ukrainian, Vietnamese
• Display modes — docked to top, docked to bottom, or free-floating window
• Appearance — Light / Dark theme, four font sizes, adjustable background opacity
• UI language — Japanese or English; auto-detected from Windows system language on first launch
• Automatic model download — model files are fetched on first use and cached in %LOCALAPPDATA%\MoonLiveCaptions\Models\
• Session recording — audio and transcript are saved to a Recordings\ folder next to the executable
• Settings persistence — all preferences saved to %LOCALAPPDATA%\MoonLiveCaptions\settings.ini
• Single-file executable — NuGet dependencies are embedded via Costura.Fody; only moonshine.dll and onnxruntime.dll need to ship alongside the exe

Requirements

Requirement	Details
OS	Windows 10 / 11 (x64)
Runtime	.NET Framework 4.8.1 (pre-installed on Windows 11)
Native DLLs	moonshine.dll + onnxruntime.dll (see below)

Getting Started

Download a Release

Download the latest MoonLiveCaptions.zip from the Releases page and extract all files to a single folder. Run MoonLiveCaptions.exe.

On first launch the application detects the Windows system language and starts in Japanese or English accordingly. It will automatically download the model for the selected transcription language (~50–150 MB depending on language).

Build from Source

1. Obtain native DLLs

The native libraries are not checked into this repository. Extract them from the moonshine-voice PyPI wheel:

pip download "moonshine-voice==0.0.49" --no-deps --platform win_amd64 --only-binary :all: -d wheels/

$whl = Get-ChildItem wheels/*.whl | Select-Object -First 1
$zip = $whl.FullName -replace '\.whl$', '.zip'
Copy-Item $whl.FullName $zip
Expand-Archive -Path $zip -DestinationPath wheel_contents -Force
Copy-Item "wheel_contents/moonshine_voice/moonshine.dll" lib -Force
Copy-Item "wheel_contents/moonshine_voice/onnxruntime.dll" lib -Force

Alternatively, place the two DLLs in the lib\ folder by any other means.

2. Build

dotnet restore MoonLiveCaptions.csproj
dotnet build MoonLiveCaptions.csproj -c Release --no-restore -p:Platform=x64

The output is written to bin\Release\net481\. Or open MoonLiveCaptions.sln in Visual Studio 2022 and build from there.

Project Structure

MoonLiveCaptions.csproj      WPF project targeting net481 / x64
App.xaml(.cs)                Application entry point; loads AppSettings on startup
CaptionWindow.xaml(.cs)      Main window (docking + floating logic)
Converters/                  WPF value converters
Helpers/
  AppBarManager.cs           Win32 AppBar API wrapper for edge-docking
  AppSettings.cs             Settings load/save (key=value .ini, no extra deps)
Native/
  MoonshineNative.cs         P/Invoke declarations for moonshine.dll
Services/
  AudioCaptureService.cs     NAudio-based speaker loopback + microphone capture
  TranscriptionService.cs    Moonshine streaming transcription + model download
ViewModels/
  CaptionViewModel.cs        Main view model (MVVM); manages all app state
  RelayCommand.cs            ICommand helper
  ViewModelBase.cs           INotifyPropertyChanged base
lib/                         Place moonshine.dll and onnxruntime.dll here

Data & Settings Locations

Path	Contents
%LOCALAPPDATA%\MoonLiveCaptions\settings.ini	User preferences
%LOCALAPPDATA%\MoonLiveCaptions\Models\	Downloaded model files
<exe folder>\Recordings\	Session WAV audio and transcript text files

Technology Stack

• WPF / .NET Framework 4.8.1 — UI framework
• NAudio 2.2.1 — audio capture (WASAPI loopback + microphone)
• Moonshine — on-device speech-to-text via P/Invoke (moonshine.dll)
• ONNX Runtime — neural network inference backend (onnxruntime.dll)
• Costura.Fody — embeds managed assemblies into the single exe at build time

CI / Releases

GitHub Actions (.github/workflows/build.yml) builds on windows-latest for every push to main or master. Pushing a tag creates a GitHub Release with a MoonLiveCaptions.zip containing the exe and the two native DLLs.

License

See LICENSE.

Transcription Engine

The project relies on the moonshine library for speech-to-text transcription. Moonshine is an open-source neural network framework that provides pre-trained models and a simple API for converting audio input into text. The application uses the moonshine.dll native component, which is included via the lib folder as described in lib/README.md.

For more information about moonshine, visit the GitHub repository:

https://github.com/moonshine-ai/moonshine

Building and Running

The project includes a GitHub Actions workflow (.github/workflows/build.yml) that demonstrates the full build process. It runs on windows-latest and performs the following steps:

1. Checkout the repository.
2. Use Python to download the moonshine-voice wheel from PyPI (currently pinned to 0.0.49) and extract the native moonshine.dll and onnxruntime.dll into the lib folder.
3. Setup the .NET SDK (9.x) and restore NuGet packages for MoonLiveCaptions.csproj.
4. Build the project in Release configuration targeting x64.
5. (CI-only) upload the resulting binaries as artifacts and create a GitHub release when the build is triggered by a tag.

You can follow the same procedure locally:

# clone and navigate to repo
git clone https://github.com/yukkes/moon-live-captions.git
cd moon-live-captions

# ensure Python 3.13+ is available, then download/extract DLLs
pip download "moonshine-voice==0.0.49" --no-deps --platform win_amd64 --only-binary :all: -d wheels/

$whl = Get-ChildItem wheels/*.whl | Select-Object -First 1
Copy-Item $whl.FullName ($whl.FullName -replace '\.whl$','.zip')
# older PowerShell on windows-latest uses -DestinationPath
Expand-Archive -Path ($whl.FullName -replace '\.whl$','.zip') -DestinationPath wheel_contents -Force
Copy-Item "wheel_contents/moonshine_voice/moonshine.dll" lib -Force
Copy-Item "wheel_contents/moonshine_voice/onnxruntime.dll" lib -Force

# build with dotnet (requires .NET SDK on Windows)
dotnet restore MoonLiveCaptions.csproj
dotnet build MoonLiveCaptions.csproj -c Release --no-restore -p:Platform=x64

Alternatively, open MoonLiveCaptions.sln in Visual Studio after placing the two DLLs manually in lib, then build and run the MoonLiveCaptions project.