Turn any song into a karaoke track in one click. Turbo AI Karaoke Maker removes the lead vocal with on-device AI, can keep the backing vocals, and opens the result in a built-in Karaoke Studio with synced lyrics, live key change and a per-stem mixer. It all runs on your own PC.
Turbo AI Karaoke Maker is a Windows desktop application that separates songs into their parts using artificial intelligence. Drop a track onto the window, press one button, and a few minutes later you have a clean karaoke version with the lead vocal taken out, plus the vocal itself saved as a guide. No account, no subscription credits and no waiting in a cloud queue.
It goes well beyond a single vocal remover. The sidebar groups eleven working tools: Pro Separation, Vocal Remover, Background Music Remover, Acapella Extractor, AI Karaoke Maker, Stem Splitter, AI Drum Remover, Bass Remover, Lead & Back Splitter, Echo & Reverb Remover and Noise Remover. A twelfth and thirteenth, Piano Remover and Guitar Remover, are listed as coming soon. Two extra sections round things out: Karaoke Studio for rehearsing and exporting, and Audio Tools for converting, normalizing and key-shifting files.
The program is built for people who use the results: singers who want a backing track in their own key, DJs and producers who need acapellas and instrumentals, drummers and bass players who want to play along, podcasters and video editors who need clean dialogue, and teachers who want to isolate harmonies. It looks like a premium studio tool, with a dark interface, a waveform player and a custom title bar, and it is engineered to stay smooth while the heavy work runs.
Everything happens locally. The application window itself is titled Turbo Karaoke Studio, and the sidebar carries the promise in plain words: offline, unlimited, and your audio never leaves your PC.
The headline tool opens by default when you start the program. It creates a karaoke track and a guide vocal from each song you add, and it gives you one choice that matters: whether to keep the backing vocals.
With Keep backing vocals off, the app runs a single AI pass with the UVR-MDX-NET Inst HQ 3 model. The model produces the instrumental, and the vocal is calculated as what is left over. You get a Karaoke track and a Guide vocal. This is the quickest route and a great fit for songs where the vocal is mostly a single lead.
Switch on this option and the app takes a longer route. First the Kim Vocal 2 model isolates every vocal in the song. Then the Karaoke 2 model splits those vocals into lead and backing. The backing vocals are added back to the instrumental, so the karaoke track keeps the harmonies and sounds more like the record. The guide vocal contains only the lead.
When a song finishes, an Open in Studio button appears on its row. It loads the karaoke track and the guide vocal into Karaoke Studio as mixer channels, ready to play. The karaoke track is selected automatically and the guide starts muted, so you can bring it up whenever you want to hear how the original singer phrased a line.
Tip: Use the standard mode to preview a whole folder quickly, then rerun your favourite songs with Keep backing vocals on. The extra pass costs time, and the app says so on the option itself, but it is the better choice for songs with strong harmonies.
Karaoke Studio is the part that turns separated audio into a performance. It plays your stems together in memory, shows the lyrics as you sing, and lets you change the key without leaving the window.
A full-width waveform shows the entire song. The part already played is lit in green, and a white playhead marks the position. Click or drag anywhere on the waveform to jump there instantly. A clock shows elapsed and total time, and the display refreshes about thirty times a second.
Transport controls start, pause and stop playback. When a song reaches the end, the player resets itself cleanly so you can sing it again immediately.
Load a standard .lrc file and the Studio shows the previous line, a large current line and the next two lines. Lines highlight on time, a music note shows before the first lyric, and files that contain several timestamps on one line are handled correctly.
Move the key up or down by as many as six semitones, one step at a time, while the music keeps playing. The readout shows Original or the exact shift, such as +2 st. The tempo does not change, and at zero semitones the pitch stage is bypassed completely so it costs nothing.
Every stem gets its own channel with Mute, Solo and a volume fader that runs from silence up to 150 percent. Gain changes are ramped, so there are no clicks when you switch a channel on or off during playback.
Press Load stems and choose several files, or one finished track. Stems from this app and from other tools both work. A file named Karaoke or Instrumental is picked to play, and everything else is ready but muted.
Render exactly what you hear, including mutes, solos, volumes and the chosen key, to WAV 16-bit, WAV 24-bit or MP3 320 kbps. The file name records the key shift, for example Song (key +2), and Show in folder opens the result in Explorer. Exports are saved into a Karaoke Exports folder.
Playback mixes stems in memory on the audio thread without allocating memory, and waveform shapes are built once and frozen, so moving the playhead only redraws two cached shapes. Heavy work never touches the interface thread.
The sidebar is organized into Pro, Basic, Studio and Others. Each tool page works the same way: a drop zone, a queue, a clear summary of what you will get, and a Start processing button. The model used by each tool is shown under its name.
Creates a sing-along track in one click. By default the vocal is removed in a single pass. Switch on Keep backing vocals and the harmonies are separated from the lead and mixed back into the backing track. The lead vocal is saved separately as a guide you can bring up while learning the song.
A studio-grade four-stem split. The vocal is isolated first, then drums and bass are lifted from the clean instrumental, which means far less bleed between stems than splitting the full mix. The remainder becomes the Other stem.
Removes the lead vocal and keeps the complete instrumental. The isolated vocal is saved too, so you can use either half.
Strips music and effects so that only the voice remains. Useful for interviews, podcasts, voice-overs and video clips with a music bed you cannot use.
Pulls a clean acapella out of any song with the high-fidelity vocal model, and returns the matching instrumental alongside it.
Splits a song into vocals, drums, bass and other instruments, with a dedicated model for each stem. Pro Separation is cleaner, but this is a solid general-purpose option.
Mutes the drums for practice tracks, or extracts the drum stem on its own. Drummers can play along to a version without the original part.
Takes the bass out so you can play the part yourself, or extracts it for study.
Separates the lead singer from harmonies and backing vocals, and also gives you the instrumental. Ideal for learning harmony lines.
Removes echo and room reverb to recover a dry, clear signal. The reverb is saved on its own as well, so you can hear exactly what was taken out.
Reduces hiss, hum and room noise with an adaptive spectral gate. It learns the noise profile from the quietest parts of the file itself, then attenuates everything that does not clearly rise above it. A Reduction strength slider runs from 0 to 100 percent. It runs on the CPU and needs no model download.
Will give you a no-piano track and the piano stem on its own. It needs a six-stem separation engine that is not part of the current release.
Will give you a no-guitar track and the guitar stem on its own, using the same six-stem engine as the Piano Remover.
Every separation tool saves its parts as separate files in a folder named after the song and the tool, and every one of them can be cancelled while it runs.
Most four-stem splitters try to pull drums, bass and vocals out of the full mix all at once, and each part leaks into the others. Pro Separation works in stages instead. It removes the vocal first with the high-fidelity Kim Vocal 2 model, which leaves a much simpler instrumental. Drums and bass are then lifted from that clean instrumental, and whatever remains is returned as the Other stem.
The same staged idea powers the Lead & Back Splitter and the karaoke option that keeps backing vocals: first isolate all vocals, then divide them into lead and backing. Because each pass has a simpler job, the results are cleaner.
Three utilities in the Others group handle the jobs that come up around every karaoke project. They run on the CPU, need no AI model and save their results in an Audio Tools folder.
Convert any supported audio file to WAV 16-bit, WAV 24-bit or MP3 at 320 kbps. Choose the format with the pill buttons on the right and process a whole queue at once. It is handy for turning an old WMA or M4A collection into files your karaoke machine or phone can play.
Bring tracks to a consistent peak level so that songs in a set play at similar loudness. Set the Target peak between minus 12 and 0 dBFS in half-decibel steps. The default is minus 1.0 dBFS, which leaves a little headroom.
Shift the pitch up or down by up to twelve semitones without changing the tempo, and save the result as a new file. Use it to match a track to your vocal range before a performance, or to prepare files for a singer who is not using the Studio.
Every tool has its own queue. Drag audio files or entire folders onto the drop zone, or click it to browse. Folders are scanned for supported audio automatically, and duplicates are ignored, so you can drop the same album twice without doubling the work. Supported inputs are MP3, WAV, FLAC, M4A, AAC, WMA, AIFF, AIF, MP4 and OGG.
Each row in the queue shows the file name, a progress bar and a live status line. While a model runs, the status reports the percentage complete, the device in use (GPU through DirectML, or CPU) and an estimate of the time remaining. When a song finishes, the row lists the files that were created, and buttons let you open the output folder, send the result to Karaoke Studio or remove the row.
Press Start processing to run everything that is not done yet. Press Cancel to stop at any moment; the file in progress is marked Cancelled and the rest remain in the queue. A song that fails, for example because the file is damaged, is marked Failed with the reason and the batch carries on with the next one. Clear empties the queue without touching anything that is currently running.
The app deliberately runs one heavy job at a time. That keeps speed predictable and avoids the slowdowns that appear when several models fight for the same graphics card or memory.
Separating a full song is demanding work. Turbo AI Karaoke Maker is engineered so that it uses your hardware well and still leaves the computer comfortable to use.
The AI engine is ONNX Runtime with DirectML, which uses any DirectX 12 graphics card. If the GPU cannot be used, the tool switches to the CPU automatically and says so in the status line. A switch in Settings turns GPU acceleration off if you prefer the CPU.
Separation runs as three stages joined by small bounded channels: spectrogram preparation, model inference and audio reconstruction. While the GPU works on the current chunk, the CPU is already preparing the next one.
The app leaves one CPU core unused on purpose, so Windows, your browser and your music player stay responsive during a long job.
Loaded models stay ready between songs, so a queue of ten tracks does not pay the loading cost ten times. The app keeps up to three models in memory at once and releases them when you close it.
The separation tools use eight proven open models in the ONNX format. The program downloads each one the first time a tool needs it, shows the progress inside the queue row, and then stores it on your PC for good. About 370 MB covers everything.
The Settings and models page is the control room. It lists every model with its role, its size and an Installed or Not installed status. Download fetches a single model, Download all fetches the lot in one go, and Open models folder shows you where they live. Each download is checked against its exact file size, and an incomplete file is deleted rather than used.
If you distribute the program to a group or need a machine that never touches the internet, copy the model files into the models folder next to the program. A bundled copy is always preferred over a downloaded one, which makes a fully offline installation possible.
Choose WAV 16-bit, WAV 24-bit or MP3 320 kbps. The choice is available on every tool page, in the Studio export panel and in Settings, and it is remembered the next time you start the program. All processing is done at 44.1 kHz stereo.
Results go to a Turbo Karaoke folder inside Documents unless you change it. Change folder picks a new location, and Open folder jumps straight to it. Results are arranged by song and by tool, so nothing overwrites anything else.
The GPU acceleration switch sits on every tool page and in Settings. A line of text explains the current mode in plain language. Your preferences are saved automatically in a small settings file in your user profile.
Choosing an output format: WAV 24-bit is best when you plan to mix or master the stems later. WAV 16-bit suits karaoke players and CD-quality needs. MP3 320 gives the smallest files for phones and USB sticks.
Turbo AI Karaoke Maker is a fully local desktop application. Your songs are read from your disk, processed on your hardware and written back to your disk. Nothing is uploaded, and there is no limit on how many songs or how many minutes you process, because there is no server doing the work.
The internet is needed only for the one-time download of AI models, and not at all if you supply the model files yourself. After that, the program works on a plane, in a rehearsal room with no Wi-Fi or on a locked-down company network.
The window has a sidebar on the left and the active tool on the right. There is no project to set up and nothing to configure before your first song.
AI separation is impressive, but it is not magic, and you should know where the edges are. The Stem Splitter uses older Kuielab models: results are good but not state of the art, so choose Pro Separation when you want the cleanest four stems. Dense, heavily processed mixes may leave small traces of the removed part.
Live key change uses a phase vocoder. It sounds great within about three semitones, and artefacts become audible beyond about five. MP3 export relies on the Windows MP3 encoder, which is missing on N editions of Windows, where WAV output remains available. Piano Remover and Guitar Remover are not available in this release.
| Platform | 64-bit Windows 10 and Windows 11 |
|---|---|
| Built with | C#, .NET 8 and WPF, delivered as a self-contained application |
| AI engine | ONNX Runtime with DirectML GPU acceleration and automatic CPU fallback |
| Models | 8 models, about 370 MB in total, downloaded once or supplied offline |
| Input formats | MP3, WAV, FLAC, M4A, AAC, WMA, AIFF, AIF, MP4, OGG |
| Output formats | WAV 16-bit, WAV 24-bit, MP3 320 kbps |
| Audio processing | 44.1 kHz stereo, with mono sources converted to stereo |
| Lyrics | Synced .lrc files in Karaoke Studio |
| Internet | Only for the one-time model downloads |
| Default output | Documents\Turbo Karaoke, changeable at any time |
Yes. Every step, from decoding the song to running the AI model to exporting the mix, runs on your own PC. The only time the internet is used is the one-time download of the AI models on first use. Once a model is installed, the tool that needs it works with no connection at all, and you can also copy the model files into a models folder to ship a fully offline build.
No. The application shows the promise on its own sidebar: everything runs on this PC, and your audio never leaves it. There is no cloud queue, no upload limit and no per-minute credit system, because nothing is processed on a server.
The eight models together are about 370 MB. Each one is downloaded once, the first time you use a tool that needs it, with a progress readout. You can also open Settings and models and press Download all to fetch everything in advance. Every file is size-checked after download, so an interrupted transfer is detected and discarded instead of causing strange results.
No. A GPU is used when one is available, through DirectML, which works with DirectX 12 graphics cards from the major vendors. If no GPU can be used, or you switch GPU acceleration off in Settings, the app runs on the CPU instead. The CPU route is slower but produces the same kind of result, and the progress line always tells you which device is doing the work.
A plain karaoke track removes all vocals in a single pass, which is the fastest option. Keep backing vocals adds a second pass that separates the lead singer from the harmonies and puts the harmonies back into the backing track. It is slower, but the result sounds closer to the original record and gives the singer a more familiar arrangement to follow.
MP3, WAV, FLAC, M4A, AAC, WMA, AIFF, AIF, MP4 and OGG files are accepted. You can drop individual files or whole folders onto the window. Files are decoded to 44.1 kHz stereo before processing, so mono recordings are converted to stereo automatically.
Yes. In Karaoke Studio, press Load lyrics and choose a standard .lrc file. The current line is shown large in the middle of the screen with the line before it and the next two lines around it, and the highlight follows playback. A music note appears before the first line starts.
In Karaoke Studio you can move the key up or down by up to six semitones while the song plays. The separate Key Changer tool in Audio Tools goes up to twelve semitones either way and writes a new file. The tempo stays the same in both. Shifts within about three semitones sound very natural; beyond about five you may hear artefacts.
Yes. Each tool has a queue. Add as many files or folders as you like and press Start processing; the app works through them one after another, shows progress and time remaining for the current song, and lets you cancel at any time. Finished items can be opened in their output folder or, for the studio-capable tools, sent straight to Karaoke Studio.
Those two tools need a six-stem separation engine that is not part of the current release. Instead of showing a button that gives poor results, the application lists them in the sidebar with a Soon label. Pro Separation, Stem Splitter, the Drum Remover and the Bass Remover are fully working today.
MP3 output uses the MP3 encoder built into Windows. A few Windows editions, the so-called N editions, ship without the media components that include it. If that is your case, choose WAV 16-bit or WAV 24-bit as the output format instead; both work on every edition.
By default in a Turbo Karaoke folder inside your Documents folder. Each song gets its own folder, with a sub-folder for each tool you ran on it. Audio Tools results go into an Audio Tools folder and exported studio mixes go into Karaoke Exports. You can pick a different location in Settings or from any tool page, and open the folder with one click.
64-bit Windows 10 and Windows 11. The application is built on .NET 8 and is published as a self-contained package, so you do not need to install a separate .NET runtime.
Yes. Press Load stems and select several audio files, or a single finished track. Each file becomes a channel in the mixer. If one of the files is named Karaoke or Instrumental, it is selected to play and the rest start muted, ready for a single click.
Offline, unlimited and private. Turbo AI Karaoke Maker for Windows 10 and 11.