Makes sense! You can enable mini mode (top right) so that you can sing without the pitch track and program at the same time :D... then listen back to your take along with the original and see how you're tracking score wise.

Agree, a good teaching tool is one that understands the scaffolds are only temporary and is constantly seeking ways to reduce the presence of scaffolds as learners increase in skill. Most tools don't take on a pedagogical lens so essentially view the user's skill as static, and thus seek to maximally reduce friction because that is what is assumed to add value. This often comes at the expense of the learner.

A more generalized view is that the design of software should aim for some optimal level of friction between the person and the task (a two-way optimization), rather than simply try to diminish it (one-way).

Also, this is a neat product :)

I feel beyond being able to play precise rhythms and pitches, there is great value in being able to be off-grid in a meaningful, purposeful way.

While it is useful to be able to be precisely on grid if you want to be so — be it with the timing of your notes or the pitches played — many musicians I know are so overly fixated on the metrum of click track or the accurate pitch the forgot how to feel the ebs and flow that comes with great music that is just played in its own space and time. And pitch is no different here, sliding into or out of pitch of being able to sing/play slightly of grid is a very powerful skill.

But technique as the ability to pull it off is one thing, the knowledge when to use it and how much only comes with years of practise.

Nowadays many beginning musicians tie themselves so much to these grids, they can't imagine playing without them. To quote a musician that was asked by the sound engineer to play without click to make the track less mechanical: "That is allowed?"

I had an idea some time ago for a pitch training app that works aurally rather than visually. Basically it's a reverse auto tune. You wear headphones and hear your pitch errors exaggerated, forcing you to sing more accurately.

Not played Rock Band myself only some Guitar Hero.

https://www.themusiclab.org looks cool, I'll give it a try today when I get a chance.

Thank you. I run Little Snitch myself so I can see what's talking out.

Great. Let me know your feedback here, in the GitHub as an issue, or on Discord!

Thanks!

OnKey only listens to what you play through your other apps, then it uses what it's heard to learn the notes and then discards the song and never saves it to disk.

This means that you need to let it listen to the song the first time it plays. You can sing to that song blind (no pitch track) and it will still score you at the end.

What I do is let it listen while a playlist plays through, then when I re-listen it lines up the pitch track with the music as it plays through my music source again.

IANAL, but I think that by listening to the music that people already have access to through their subscriptions, not distributing any music, lyrics, or learned melodies i.e. keeping everything for personal use only (see https://github.com/dsmurrell/onkey#using-onkey-fairly), rights aren't required.

I might do the ranking thing later but this brings some form of centralisation to the project which it currently doesn't have, and I haven't worked out how to stop people gaming that.

I'm not great at any of these things. I do agree with you... do you have links to good resources in this regard? Maybe I can start by linking to them, and then later building the features that make sense into the app.

You probably cannot imagine how bad some people are at singing. I have people near me who sing completely out of tune, so much that it hurts. I am not an Apple guy, so I cannot test the app myself, but I will definitely send the link at least to those who know themselves that they are terrible at singing, hoping that Christmas will be less painful for my ears this year.

I've been doing classes for more than a year with an online teacher and it's been amazing and when watching the demo I saw so many of the mistakes I've done in the past and yet the app keeps saying to the singer GREAT and PERFECT but the coordination is all over the place, not to mention an octave lower and in a very monotone way (lack of prosody or emotions).

And it makes sense, you can't add emotions until you cover some of the basics.

This app can work really well once you have breath and placement nailed down and you just want to focus on pitch, but it feels like it can me misleading to people to think they sing good.

This.

As an amateur musician who can afford private lessons, I cannot understate how important a teacher is in music learning. You can try to learn things on your own and mess with these stuff, but a teacher goes a long way. They spot your mistakes instantly and help you progress efficiently.

I understand that it is not financially viable to everyone, but seriously, if you are able to, get some professional guidance.

> I never dare to try if not alone

Yes, this was me. My confidence has improved through practise to a point where it only half killed me to post that demo video. My family still ask me to relocate when I'm doing this around them... so I think you can hit the pitches and still not sound amazing.

Unfortunately, the way it works, and to avoid any copyright issues, it can't really operate on phones because you can't listen to other apps on them. Maybe the only way is to share your learnt pitch tracks (trained on the laptop) with a mobile companion app.

Discussed here:

Turn your singing voice into printable notes (in the browser) - https://news.ycombinator.com/item?id=48900686 - July 2026 (26 comments)

Thank you, I'll take a look.

really cool! Thanks for sharing

Each song has a difficulty rating from 1 to 5. You can use this to filter your library.

And you can track your scores over the takes you did of a sing in your history. So I think OnKey has you covered on these.

Let me know if it matches what you expect when you try it out.

Yep, this would have taken me maybe a few months (possibly a year) before, learning what I needed to know, and AI brought it down to about 10 days. Also have an ML background but this didn't come into play building this, I got my agent to do the research on what tools were available and which I could use based on their licenses, then created an eval set and tried the tools against one another until I landed on the best set.

Credits for the libs I used are here: https://singonkey.app/docs/credits

I don't understand either what, technically, "off-key" means. I mean, sure, in well-known songs the singer needs to sing certain parts at a certain pitch, and it sounds awful when they don't. But say this is a song you are composing. There, you have your chord progression, but the chords and their notes are some sort of guidance. You may compose the song in such a way that the singer sticks strictly to the notes in the chord as much as possible, but that sounds weird and requires no small amount of singing skill. Or you may keep the singer a perfect fourth or a minor third away from some notes of the chord for part of the melody, "off key", and it may still sound good.

Agreed, but if you're singing to the song, then you'd have to match the vocal pitch of the singer otherwise there would be a clash.

Some other singing apps allow you to transpose the entire song but it's easy for them because they play the music. In OnKey's case, it would have to silence the music from the player then replay it shifted up or down. So this might be possible, I'd have to do some experimentation.

The Mac only parts are in https://github.com/dsmurrell/onkey/tree/main/src-tauri/nativ.... I believe that Linux would need PipeWire and MPRIS in their place.

> One of my methods for my pet projects is just to wait for someone else to do it, worked this time :)

Happy to be of service!

PRs welcome.

For the user's pitch OnKey uses CREPE-tiny (https://github.com/marl/crepe) but ported to Rust with the model running on ONNX Runtime. For the singer's pitch also CREPE but the full version.

The vocal separation for the song uses UVR: https://github.com/Anjok07/ultimatevocalremovergui

You can find more details here: https://singonkey.app/docs/how-it-works

Nice :)

Noted, I haven't done a feasibility test yet, but I do have an issue up here to vote on: https://github.com/dsmurrell/onkey/issues

It's got 3 thumbs up but no indication of which OS they're from... maybe I need two issues.

It handles some mild vibrato because it's quite lenient on scoring you and if you're off pitch a bit you get full score: https://singonkey.app/docs/scoring

I'll bear this in mind though when I add the harsher scoring modes.

I have thought about this for my children who I find hard to understand sometimes :) I'm not sure what the right approach is here, I have some ideas but I'd have to test them to see if they work out.

Will do, check out https://github.com/dsmurrell/onkey/issues/1 to register your interest in other platforms. Need to test feasibility.

Not at the moment but it would be a good feature, I'll try to see what's possible here.

Thank you.

I suspected that extracting played notes out of songs might be more difficult than vocals, but my agent is telling me that it might be possible given its context around the models the app is currently using. I'll experiment with this.

It might only work for single note instruments.

Works with other languages. Personally, I love this song: TRINIX - Vaitimbora feat. Mari Froes

But find it hard to sing to as I don't speak Brazilian Portuguese so it's hard to match the words.

I will ask my friend if it works for him as he speaks the language.

I love that you love this!

You're not wrong. I do have a coding background which helps but I'm new to Rust. It started in Python with a web frontend so it somewhat evolved into this.

> why is it using webview/tauri?

Using these makes it easier to port to the other platforms later if I choose to.

About 365 of the 420 MB is the AI models which I decided to bundle so that the app wouldn't have to download additional things after the install.

My agent tells me that writing in native Swift would save (estimate) about 100–170 MB, out of about 1.4 GB when the models are active. And then I'd have to maintain the interface in two or three ways if I decided to make versions for other platforms.

If it does grow though and there are a lot of Mac users, this might be worth it though as you suggest. But I have a lot of project that go precisely nowhere so I always keep this in mind :)

In a small defence to webview/tauri, it's pretty smooth and fast, and multi platform free of charge. I'd have done this in Tauri too, but that's just my voice. But I otherwise totally agree with you.

Yes, I got this from a friend... "Lol, I could t tell the difference between you and Tony Kiedis!"

I'm terrible at detecting sarcasm usually, but this one was easy for me.

I had to look it up and maybe it helps others understand: Shaquille O'Neal was bad at shooting threes and only made one three-pointer in his entire professional career (he attempted 22).

At least they're not using him to learn lyric writing.

I now realize this comment may read much more mean-spirted than I intended--not my intention!

This is a great and cool app. I am just joking. Sorry to be an ass.