Isolate the Main Vocal

How to isolate the main vocal

1. Upload the file. MP3, WAV, M4A or FLAC, up to 10 minutes. Sign in with Google the first time.
2. Press Isolate the main vocal. The button shows the price first: 1 credit for every minute. New accounts start with 5 credits. Most files take 1 to 3 minutes.
3. Listen and download. You get Voice 1 and Voice 2. Click a reel to hear one alone. Each downloads as an MP3 at 320 kbps. Failed jobs are refunded.

How it decides what is "main"

The model looks for the loudest voice and tries to pull it away from the other voices. It does not understand who the lead singer is. The loudest voice in the mix is the one it goes after.

The files are labelled Voice 1 and Voice 2, not "main" and "other". Listen to both and see which holds the voice you wanted. In most cases it is the clear, loud one.

Where it is useful

A lead singer with a crowd of backing voices. Pull the lead away from the stack to learn the melody or write out the lyrics.

Speech over other speech. A recording of several people talking, where one voice is clearly closest to the microphone, can sometimes be split so the near voice is easier to hear.

Sampling. A producer wants one singer from a gang vocal.

Transcription. Listening to a loud voice alone is easier than picking it out of a pile.

What to expect

It works best when one voice is clearly louder than the rest. When several voices are about equally loud, the model has to guess and may mix them between the two files. Music under the voices can also confuse it, so a vocal-only file gives a cleaner result. If you only have the full song, run the Acapella Extractor first.

For a defined lead and backing stack, Separate Lead and Backing Vocals is built for that job. For two singers of similar loudness, try Separate Singers.