Sing!
Pitch training for karaoke singers
Hear how you'll sound. See how far you are from it.
Pick a phrase from a song you know. Sing it.
Sing! plays you two takes of the same words, in your own voice, at your own tempo: the one you sang, and the one with the pitch quietly corrected. Nothing else changes — not the timing, not the timbre, not the words. The only difference you hear is the thing you are working on.
Then it shows you the same difference as a picture: where you were flat, where you were sharp, and by how much. The exercises come out of that song, not out of abstract scales.
How it works
Your own songs
You bring the song you actually want to sing: upload a recording and the server separates the voice from the backing track, so you can hear the vocal line on its own and work against it.
Compared with the singer, not with a grid
Most apps score you against a row of notes on a piano roll. Sing! lines your take up with the contour the singer actually sang — the slides, the scoops and the held notes included — because that is what you hear when you listen to the record.
An honest measurement
The score is computed on your untouched voice. The corrected version is something you listen to, not something that quietly improves your numbers, and the takes you sing in exercises are never stored.
Roadmap
Working today
- Bring a song as a file; the voice is separated from the backing track on the server
- Listen to the isolated vocal line, trim what you do not need, shift the key to fit your range
- Record a reference in your own voice by singing along with the vocal line
- Hold-a-note exercise with a corridor, built on the first note of your song
- Sing a whole phrase and hear three versions of it back
- Pitch detection checked against a public dataset of real, labelled singing
Planned
- Web release planned for 1 November 2026
- Exercises built from the phrases of your own song, not from scales
- A phone app after the web version
Request early access
Sing! is in closed beta. Early requests get in first and shape the release.
Frequently asked questions
Do I need a good ear for this?
Most likely you already have one. In an online study of 632 adults, 11 per cent sang inaccurately by a criterion of 50 cents on 90 per cent of notes (Pfordresher & Demorest, 2021). Earlier work found that most inaccurate singers hear the difference perfectly well: what is broken is the link between ear and voice, not the ear itself (Pfordresher & Brown, 2007). Sing! is built for that link.
What happens to my recordings?
The song you bring and the reference you record are your working material: they are kept on the server so you can come back to them, and you can delete them. The short takes you sing in exercises are different — they live in memory for the length of the request and are never written to disk, to a log or to analytics.
What do I need to run it?
A browser with a microphone. The web version comes first and is built for a phone screen; a phone app comes after it. For recording your reference you need headphones, because the microphone would otherwise pick up the vocal line from the speaker and mix two voices into one take.
How is this different from Smule or Yousician?
Apps that make your voice sound better do not measure how you sang, and apps that measure how you sang do not touch your voice. Sing! does both, in that order: it shows you your own corrected voice as the target, and it measures the distance to it on the untouched take.
Does the app teach you to sing?
We do not claim that. What is shown in the literature is narrower and we stick to it: in a study of 75 untrained adults, twenty minutes of practice improved accuracy significantly only in the group that got visual feedback on pitch, with an effect size of d = 0.49 against the control (Berglin, Pfordresher & Demorest, 2022). The same body of work also shows that constant feedback becomes a crutch, which is why the hints in Sing! are designed to fade.
When is the release, and how do I get in before it?
The web release is planned for 1 November 2026. Until then Sing! is in closed beta and needs an invitation code. Leave your email below and we send you a code when a place opens up. Places open in batches, because separating a song into voice and backing track takes several minutes of one machine.