Questions, answered
Still stuck? Email support@beltr.app, and you’ll hear back from the person who builds Beltr. The support page has other ways to get in touch.
Getting started
What is Beltr?
A desktop karaoke app that turns songs you already own into karaoke: the vocals removed on your own computer, the lyrics on your TV, and phones as wireless remotes and mics.
How does it work?
Add a song and Beltr’s AI separates the vocals from the instruments in a minute or two, on your computer. Synced lyrics appear on your TV or monitor, and guests join from their phones by scanning a QR code to queue songs and control playback.
Does Beltr come with any songs?
Yes, one: the Beltr theme song, ready to play the moment you open the app, in two versions. One is the AI path (vocals separated, word-synced lyrics, a music-video background and live pitch scoring); the other is a classic CD+G version, rendered the way a hardware karaoke machine shows it. Both step aside as soon as you import your first song.
How do I get Beltr on my TV?
The most reliable way is an HDMI cable from the computer running Beltr to the TV: no lag, the best audio, and pitch scoring works right away because the mic is on the same computer. You can also open Beltr in your smart TV’s browser, use AirPlay (Mac to Apple TV) or Chromecast, or install the free TV app: Beltr Client for Apple TV and for Android TV and Google TV.
What are the system requirements?
macOS 12 or later on Apple Silicon (M1 or newer), Windows 10 or 11, or Linux. 8 GB of RAM at minimum (16 GB recommended), a modern multi-core CPU and about 2 GB of free disk space. Intel Macs are not supported.
Is the Mac app notarized by Apple?
Yes. The macOS build is signed and notarized by Apple, so it opens normally with no Gatekeeper warning. Drag it to Applications and launch it.
Can I run Beltr on Unraid or a home server?
Yes. Beltr is in the Unraid Community Applications store: open the Apps tab, search for Beltr and install one of the three templates. Beltr runs on the CPU anywhere, Beltr-NVIDIA is the CUDA build for servers with an NVIDIA GPU and the Nvidia-Driver plugin, and Beltr-Intel is the OpenVINO build for Intel integrated graphics and Arc cards. Install one, not all three; AMD GPUs are not supported in the container. The self-hosting page covers Docker Compose, TrueNAS and Apple Silicon Macs.
Does it work offline?
Yes. The internet is needed to install, for a one-time download of the alignment model and to activate your license key once. After that, vocal separation, lyric alignment, playback, scoring and the phone remotes all run on your own machine. Looking up lyrics online and music-video backgrounds are the only features that use the internet, and Beltr falls back to transcribing on your computer when lyrics can’t be fetched.
Songs and formats
What audio formats are supported?
Any MP3, FLAC, WAV or MP4 you own goes through vocal separation. Karaoke files play as they are: CDG and MP3+G discs, .kar and .mid files, and Thai NCN songs, with lyrics and timing read from the file.
Can I import my existing CDG or MP3+G files?
Yes. Add an MP3+G or CDG zip and it plays straight away, next to your AI-made tracks in the library. Disc tracks carry an MP3+G badge so you can tell them apart.
Can I play Beltr tracks on a karaoke machine?
Yes. Beltr exports any prepared track as MP3+G, CDG or .kar. Burn it to a disc, copy it to a USB stick, or load it onto a hardware karaoke machine that reads those formats. More on the MP3+G page.
What’s MP3+G?
The standard karaoke-disc format: an MP3 of the music paired with a CDG file that paints the synced lyrics on screen. It’s what commercial karaoke discs and karaoke machines use. Beltr imports and exports it.
Do music videos play behind the lyrics?
Yes. When a song’s official music video is available, Beltr can play it on the TV behind the words, dimmed under the lyrics. It’s optional and runs as you add songs. Fetching a video takes two free, open-source tools that Beltr doesn’t bundle: yt-dlp and a JavaScript runtime, Deno (or Node.js if you already have it), which YouTube now requires. The easy way is to let Beltr fetch both: Settings, then Music-video backgrounds, then Install. It checks each download against its publisher’s checksums and keeps its own copy of yt-dlp up to date.
Can I record and share my performance?
Yes. Reveal Reel records your take as you sing and makes a vertical video clip. After the song, scan the QR code on the screen to open the clip on your phone, then share it or save it. The clip stays on your machine; nothing uploads on its own.
Can I use songs from streaming services?
No. Beltr works with audio files you own. It doesn’t stream from or connect to any streaming service.
How it works
How long does vocal separation take?
A minute or two per song on most computers: quicker on a recent Apple Silicon Mac, longer on an older PC that only has a CPU. The separation model ships inside Beltr, so there’s no download before the first song; the first song whose lyrics need word-level timing fetches a one-time alignment model.
What AI does Beltr use to separate vocals?
Beltr runs UVR MDX-Net, an open-source vocal-separation model, on ONNX Runtime. It splits a song into the vocal and the instrumental on your own computer, and you keep both stems. Nothing is uploaded. Versions before 1.60 used Demucs v4, which is why older reviews and release notes mention it.
How are the lyrics timed to the music?
Beltr finds or accepts the lyrics, then force-aligns them to the vocal track, word by word. Synced lyrics come from LRCLIB when they exist, and you can also paste or import your own. When no lyrics can be found, Beltr can write a draft transcription on your computer with an on-device speech model before aligning it. That step is optional and never runs when lyrics were found.
Does Beltr use my GPU?
Apple Silicon Macs are accelerated out of the box. On Windows and Linux, an optional GPU pack in Settings speeds up vocal separation: CUDA for NVIDIA cards, and DirectML on Windows for AMD and Intel graphics where the driver supports it. Nothing else needs the GPU, and Beltr always works on the CPU alone.
Why does separation quality vary?
Studio recordings give the best results. Live recordings, tracks with heavy effects and low-bitrate files can separate less cleanly. If a vocal isn’t fully gone, you can still lower it and sing over a guide.
Phones
How do phones connect?
Guests scan the QR code on the TV. It opens a page in their phone’s browser, with no app download and no account.
How many phones can connect?
There’s no hard limit. Every connected phone can browse the library and queue songs.
Do guests need to install anything?
No. The remote works in any modern mobile browser, with nothing to download or sign up for.
Is there a phone app?
The remote needs no app. For singing, Beltr Remote is a free app that turns your phone into the wireless mic, with live pitch scoring. It’s on the App Store and Google Play, and it needs the Beltr desktop app.
Price and license
How much does Beltr cost?
$19.99 once. There’s no subscription. For the arithmetic against a monthly plan, see how KaraFun’s monthly price compares with Beltr’s.
What’s in the free trial?
The full app, with 5 song separations. No card and no time limit.
What happens after my 5 free songs?
The app keeps working, and you can play every song you’ve already prepared. It won’t prepare new songs until you buy a license.
Can I get a refund?
Yes. If Beltr isn’t for you, email support@beltr.app within 14 days of buying for a full refund, with no forms and no questions. Details are on the refund page.
How many devices can I use?
Each license key works on 2 devices.