CD+G karaoke, generated
Every track keeps its ghost.
Upload an MP3. GhostTrax lifts the vocals off it, times every word, and hands back a CD+G package your player already opens — plus the isolated vocal, so the original singer can be faded back in when the room needs it.
A CD+G screen with a vocal-guide mixer
Playback mix
vocal 0
Instrumental only. The singer carries it — the way karaoke has always worked.
The part nobody else ships
The original vocal comes with the track.
Separating a song produces two halves, and most tools throw one away. GhostTrax keeps both. The isolated vocal rides along in every package, so it is there when you need it.
Rescue a singer mid-song
A nervous singer loses the melody in verse two. Ease the original vocal up on the GhostTrax player and they have something to hold onto, then ease it back out.
Check the timing before it goes in the book
Play the vocal against the graphics you just generated. If a word lands early, you will hear it immediately instead of discovering it in front of a room.
Run it as a guide track
Bring the vocal in low under a difficult song and let it sit there all the way through. New singers stay on pitch, and the room never hears a train wreck.
Nothing extra to generate
The vocal stem is a by-product of separation, not a second job. Every track you make already has one waiting in the zip.
One job, one zip
What lands in your library
Four files, named the way your player expects. The MP3 and the CDG share an exact basename, because that pairing is how most CD+G players find the graphics for a track.
Three passes
How a track gets made
Every job reports these stages as it runs, so you always know where a track is. A song takes a few minutes end to end.
separating
The voice leaves the body
A source-separation model splits the recording into a vocal stem and a backing track. One becomes your instrumental; the other becomes the ghost.
transcribing
Every word gets a timestamp
Speech recognition runs on the isolated vocal, not the full mix, so the words land where the singer actually sang them.
rendering
The graphics get written
Those timings become a CD+G packet stream at 300 packets per second, each word lighting up as it comes around.
Host software
Auto-KJ is the only hosting software with the ghost built in.
The vocal stem is only useful if your host software can reach for it without stopping the show. Auto-KJ is the only karaoke hosting software with GhostTrax vocal restore integrated — the fader is on your console, next to the rotation, not in a second app you have to alt-tab into mid-song.
Its Ghost-Trax Renderer plays zipped MP3+G packages straight onto the singer's screen, so a track you generate is playable the moment it lands in your library.
Ghost-Trax Renderer
High-performance CDG rendering engine. Plays zipped MP3+G files instantly on secondary monitors with pixel-perfect accuracy.
Fair Slot Engine
Assigns 1–4 singer slots per song and places requests in the earliest open round for all members, preventing rotation manipulation.
Smart Unpause
Singers can pause requests. On unpause the system slots them at position #2, protecting the current and announced "on deck" singers.
Duplicate Cooldown
Prevents immediate repetition of the same track — configurable as a session-long ban or a custom hourly cooldown.
Local LAN Fallback
If the internet fails, the host app spawns a local HTTP server and WebSocket relay. Singers keep requesting over local WiFi.
Duets & Bands
Bands enable direct co-op requests, and the Duet Finder compares catalogs between friends to recommend shared favourites.
GPS Show Finding
Singers discover nearby active shows by geolocation, with room codes as a fallback in low-signal bar basements.
Auto-KJ Normalizer
Analyzes and normalizes track volume, keys and tempos automatically. Included free with all license tiers.
Details that matter at a show
It fits the way you already work
Your naming convention survives
Hand it KV12345 - Artist - Title.mp3 and every file comes back under that same name. Nothing to rename before it goes in the library.
Plain CD+G, no player lock-in
The output is a standard .cdg packet stream beside a standard MP3. It works in the software you already run.
Lyrics in text, too
Every job includes a plain-text lyric dump — useful for fixing a line the model misheard, or for searching your library later.
Built on Cloudflare
Jobs run in the cloud, not on your show laptop. Upload from the rig and the package is waiting whichever machine you open next.
Tracks are yours to keep — but download them
We hold a finished track for 7 days, and delete it 24 hours after you download it. Your account shows a live countdown next to every track. This is storage, not a library — put the files in your own library and they are yours forever.
Pricing
Pay for tracks, not seats
Every plan returns the same four files, ghost included. No watermarks. Go over your monthly tracks and each extra one is $0.99 — no plan change needed. Finished tracks stay on our servers for 7 days, or 24 hours after you download them, whichever comes first — the files themselves are yours to keep for good.
Working KJ
$19.99
per month · 20 tracks
- 20 tracks a month
- Extra tracks $0.99 each
- Everything you make is yours
Venue
$99
per month · 150 tracks
- 150 tracks a month
- Extra tracks $0.99 each
- API access for your own tools
Someone is going to request it tonight.
Upload the song now and it will be in the book — ghost and all — before their turn comes around.