About this site
| Site name | Karaoke Trainer |
|---|---|
| URL | https://ktkwmr.com/karaoketrainer/ |
| Operator | ktkwmr (individual developer) |
| What we offer | A free web app for karaoke practice: it separates the vocals of your own song file with AI on your device and shows your pitch over the original melody as you sing, with key change and a key suggestion for your vocal range |
| Contact | GitHub Issues |
Contact
For questions, bug reports, and suggestions, please open an issue on GitHub Issues. A GitHub account (free) is required.
Issues are public. Do not include personal information, and do not attach songs or recordings. If something fails, the error message, the length and format of the file (for example 4:30, MP3), your browser, your device, and the headphones you use are enough.
Disclaimer
- The separation is done by an AI model and is not perfect: some of the accompaniment can stay in the vocals and the other way round. The melody line and the judging of your pitch are made automatically and can be wrong, for example with backing vocals, harmonies, or spoken parts. They are a guide for practice, not a score or an evaluation of your singing. We do not guarantee the results or that the information on this site is correct or complete.
- Separation needs memory and processing. On some devices it can be slow or fail, and the device can get warm.
- Use song files you bought or recorded, for your own practice. Publishing or distributing separated audio of other people's songs may infringe copyright. You are responsible for how you use the files.
- We are not responsible for any loss or damage resulting from the use of this site.
- Links to other websites are provided for convenience. We are not responsible for their content.
- The content of this site may be changed or removed, and the service may be stopped, without notice.
Copyright
- The app's source code is published under the MIT License. The libraries used by the app are listed under Settings → About in the app.
- The AI model is Spleeter (2 stems) by Deezer, whose code and trained models are published under the MIT License, converted to ONNX by the sherpa-onnx project (Apache License 2.0). This site serves the model files itself. The licenses of the model and the libraries are listed under ⚙ → About in the app.
- The text and images on this site are © 2026 ktkwmr.
Privacy
How this site handles information is described in the privacy policy.