How does an AI voice music separator work?
Older vocal removers cancelled out whatever sat in the middle of the stereo mix, which also took out bass and drums and left a hollow sound. An AI vocal separator works differently: a model trained on hundreds of multitrack songs recognises what a voice sounds like and splits the song into the vocals and everything else. Nothing is cancelled out, so the instrumental keeps its full width and low end. The same model can isolate a voice from the audio of a video.
Instrumental or acapella?
Choose Remove vocals to get the instrumental — the backing track for karaoke and covers. Choose Keep vocals to get the acapella — the voice on its own for remixes. Choose Both and you get the two files, which costs the same.
Getting the cleanest result
- Start from the best copy you have. The voice music separator can only work with what is in the file. A WAV or FLAC keeps more detail than a low-bitrate MP3 or audio ripped from a video.
- Heavy reverb leaves a trace. When the vocal has a long echo baked in, a faint ghost of it can stay in the instrumental.
- Backing vocals count as vocals. Harmonies and choirs are removed together with the lead singer.
- Trim first. Credits are counted per minute, so cut the song to the part you need.
Using the tracks you make
Running a song through a voice music separator is fine for practice, karaoke at home and private remixes. Publishing a remix or a cover built on someone else's recording usually needs permission from the rights holder — the separated tracks are still their work.