Audio Engine · v0.6.0

A local audio AI engine, on your machine.

Speech-to-text, text-to-speech and more — powered by audio.cpp, a pure-C++/ggml engine. Install it once; it keeps a local server running and puts an OpenAI-compatible API on 127.0.0.1. No Python, no account, nothing uploaded.

Download Source

WhisperKey's heavyweight sibling: where WhisperKey is a tiny, single-purpose dictation tool, this is the full engine for when you need one.

What it does

Speech to text

Transcribe audio locally — SenseVoice, fast and accurate, with punctuation.

Text to speech

Natural neural voices with Pocket-TTS; more voices and cloning available.

Voice cloning

Clone a voice from a short reference clip (via the underlying engine).

audio.cpp also brings diarization, VAD, source separation and music — room to grow.

Use it right in the app

No code needed — the app has a built-in panel for each job, all running on your machine.

Speak

Type or paste text, pick a voice and speed, and get a voiceover you can play and save.

Transcribe

Record from your mic or drop in an audio file and get the text back — recording is native, not the browser.

Podcast

Paste notes and split them into a back-and-forth, assign two host voices, and render one stitched show.

Self-contained

The engine ships inside the app. The model weights (~390 MB) download themselves the first time you run it, after which it works entirely offline. Nothing to install alongside it, no toolchain, no venv.

Use it from anything

Any app on your machine can call the OpenAI-compatible API
Your LLMs can drive it through the bundled MCP server
Other apps (like diy-platform) can attach to it instead of shipping their own engine

# LLMs: point an MCP client at
node mcp/server.mjs        # AUDIO_ENGINE_URL=http://127.0.0.1:8080

# Any app: the OpenAI audio API
curl :8080/v1/audio/speech -d '{"model":"tts","input":"hello"}' -o out.wav

Download

Linux

v0.6.0

x86_64 · GNOME, KDE or any X11/Wayland desktop

.deb package 71 MB AppImage 139 MB

.deb package — Debian, Ubuntu, Mint, Pop!_OS
AppImage — Any distribution, nothing to install

Windows

v0.6.0

Windows 10 1803 or newer · 64-bit

Installer 39 MB

Installer — Per-user, no administrator prompt

macOS

Not yet built

A macOS build is on the way — not yet packaged for download.

SHA-256 checksums

Verify with sha256sum on Linux or Get-FileHash in PowerShell. The builds are unsigned, so this is the only way to confirm you have the file that was actually published.

AudioEngine_0.6.0_amd64.deb	083131cf64ad5a66081186049d0ac6c327c88980be83f64f37759a77bba735f0
AudioEngine_0.6.0_amd64.AppImage	93f5fc1799cb84c43c187265fe3cbd029b66f0bddd643fc8a8406485203e7c69
AudioEngine_0.6.0_x64-setup.exe	23aaecb54cd5ad550fdb9e78f420d2c6ac11c4f42fa63870086a075cd9ee02c9

The builds are unsigned; on first launch your OS may warn, and every download has a published SHA-256 above. The Windows installer is cross-compiled from Linux — SmartScreen will warn (More info → Run anyway).

Prefer to build Windows yourself?

The Windows download above is prebuilt. To build it natively instead (~15 min), against an audio.cpp checkout:

# PowerShell
irm https://dl.vibedout.app/audioengine/build-windows.ps1 -OutFile build-windows.ps1
$env:AUDIO_CPP_SRC = "C:\path\to\audio.cpp"
powershell -ExecutionPolicy Bypass -File build-windows.ps1

Direct link: build-windows.ps1. It installs any missing tools, builds the engine with MSVC, and produces the same installer.