Vocal upscaler

Drop a vocal recording to restore, or select a file. It opens in the vocal upscaler, ready to run.

WAV · AIF · AIFF · MP3 · M4A · FLAC · OGG

How to restore a vocal recording

A filter can only take things away from a bad vocal: some hiss, some room. It cannot put back the top end a phone microphone never recorded. This does something different. It listens to the take, the words, the notes and who is singing, and sings it again, clean.

  1. Add your vocal: Drop a WAV, AIFF, MP3, M4A, FLAC or OGG of a sung vocal on its own. From a full song, isolate the vocal first.
  2. Press Restore vocal: The recording is heard for its words, its melody and its singer, then sung again on the vocal model in that voice.
  3. Play it and download: The restored vocal comes back as a WAV the length of the take. Play it against the original before you keep it.

More ways to work with your music

Try our online tools to separate, transcribe and convert. Doseedo does it all.

Questions about the vocal upscaler?

We have answers.

Is this noise reduction?

No. Noise reduction filters the recording you have. This sings the take again in the same voice, so the noise and the room are not in the result at all.

Will it still sound like the singer?

It is sung in the singer’s own voice, read from the recording. In our test the likeness to the original singer scored 0.86 on a 0 to 1 scale. Listen against the original: a re-sung take is close, not identical.

Does it work on a full song?

It takes a sung vocal on its own. For a finished song, pull the vocal out with the acapella extractor first, then restore it.

Does it work on speech or rap?

It is built for sung vocals. A take it cannot read as a sung vocal is re-rendered plainly instead, and the result says so.

Is it free?

Yes, to start. A free account comes with 120 credits a month, with no card needed to start, and a render is 10 credits for every 30 seconds of vocal. Pro has 2,000 a month.

Try our free vocal upscaler

  • Re-sung, not filtered

    The result is a new performance of the same take, not a processed copy of it. That is why it can have high frequencies the recording never had, and why the room and the noise are simply absent rather than reduced: they were never part of what was sung again.

  • What it keeps

    The singer’s voice, the words as they were sung, the notes, and the phrasing and timing of the take. It reads the recording’s own structure and holds the new vocal to it, so the restored take lines up with the original and with the track it belongs to.

  • What we measured

    Four clean takes were made roomy, noisy and band-limited to 8 kHz, restored, and scored against their clean originals. Speech-quality score (PESQ) came back at 3.08, against 3.39 for the clean take itself and 1.33 for a plain re-render. The top of the spectrum, above 8 kHz, came back to within 2 dB of the original.

Vocal upscaler, vocal enhancer, vocal restoration: one job

A vocal upscaler, an AI vocal enhancer, vocal restoration and a way to fix a noisy or lo-fi vocal recording all name this one job. To get the vocal out of a song first, use the acapella extractor; to change what it sings, the lyric changer.

Try these Doseedo tools