Music & sound

Feedback you can hum

A song is unfinished for far longer than it is finished. It starts as a voice memo hummed into a phone on a walk, becomes a rough bounce sent to a friend at midnight, grows through a dozen mixes that are each nearly right, and ends — months later, sometimes years — as a master somebody finally says yes to. Almost the entire life of a piece of music is spent in that in-between, being passed around and listened to and talked about. And the strange thing is that the tools for talking about it have always been the wrong size. At one end sits the full DAW session, which is a workshop when what you wanted was a napkin; at the other sits the voice memo recorded over a phone speaker while the track plays somewhere in the room behind it. By the time you have set up something proper, the feeling that made you reach for the idea has usually left the building.

So the first thing we built for music is a scratch pad. When a collaborator drops a demo into your shared Space, you can open it right there in the browser, scrub to the spot where you hear something, and press Record an idea. Headphones in, the track plays from the playhead, and you sing or play your part over the top. It loops back so you hear the two together; a take that misses is retaken without ceremony; and when one lands, Saveflattens it into a finished-sounding, lossless file that appears in an Ideas folder beside the track it came from — a thing you can play anywhere and send to anyone, while the original stays exactly as it was. And if the space you’re recording into isn’t yours to save into, the blend downloads straight to your machine instead, yours to keep either way. There is deliberately nothing multitrack about it. It exists for one sentence, the one that starts every good collaboration in music: what if it goes like this here?

Night Driving (demo).mp3

Record an idea from 1:26

This is a headphones feature — pop your headphones or AirPods in.

Play backRetakeSave to this space

Saved to Ideas — “Night Driving — idea.flac”

It is a headphones feature, and we say so plainly rather than pretending otherwise — the microphone should hear you and only you, which is how musicians have recorded over a track since the first pair of cans was plugged into a desk. And the ideas accrue the way sketches do in the margin of a notebook: beside the work, dated, disposable, each one a feeling caught while it was still warm.

Then the track grows up, and the conversation changes register. A mix goes out for approval, and here the ritual has barely moved in twenty years: you listen, you stop, you scrub back to find the moment again, and you type “2:13 — is the vocal a touch hot coming out of the bridge?” into an email, copying the timestamp over by hand from a player that lives in a different window. The file itself arrives through a transfer link that will quietly die in a week. The feedback ends up scattered across an email thread, a messaging app, and one phone call nobody wrote down, and three mixes in, nobody is entirely sure which bounce the notes were about.

In Spaces the waveform itself is the conversation. You see the whole track at a glance and scrub it by eye; a double-click drops a note pinned to that exact second; and dragging across a section loops it, so you can live inside the chorus for a minute while you work out what you actually think. Notes rise to meet you as the playhead crosses them. When mix two arrives it stands beside mix one rather than replacing it, each round of notes staying with the version it was about, and a note that has been dealt with is marked resolved — so the thread reads as a history of the mix getting better, which is what it is.

Night Driving — mix v2.wav
48 kHz · 24-bit · WAVLossless

Tim1:32Voice note

The low end’s lovely, but the vocal’s sitting a touch hot right through this section — maybe half a dB down and see how it feels?

Heard in full, off the real file — and if a note is deleted, the recording goes with it.

The person on the other end needs no account and no instructions. The link opens the same player, they leave their notes under their own name, replies travel back the other way, and both sides are told when there is something new. An artist can do this from a phone in the back of a van. The mastering engineer sees exactly which second of which version the note belongs to. Nobody copies a timestamp by hand ever again.

And because this is music, we did something with feedback itself that we have come to think matters more than any other detail on this page. Musical feedback is spoken and sung and hummed — the client who cannot name the frequency can hum you the problem in three seconds flat. So there is a microphone in the composer, and when someone holds it and says, or sings, what they mean, the recording itself is the note. The team plays it back and hears the hesitation, the emphasis, the melody of the thing — everything a typed-out version flattens away. A quiet transcript is kept behind an opt-in reveal so the note stays searchable, made by a speech model we run ourselves in Amsterdam that trains on nothing; and if the note is ever deleted, the recording goes with it.

Recording 0:12Stop

Your actual recording is sent — they hear your voice, not a transcript of it.

At mastering, the stakes invert. The conversation is no longer about ideas; it is about half a decibel and a limiter ceiling, and the one unforgivable thing a delivery tool could do is make a clean master sound broken. So playback is lossless end to end, and before a single sample plays, the whole file is buffered into memory — after which the network can do what it likes, because a wifi stumble physically cannot reach the audio. The pristine original sits one click away for the final critical listen on your own monitoring, where that listen belongs. And you will find no warnings here about your listening environment: the player states what the file is and trusts your ears, because a tool for professionals should behave like one.

Music has always been the most international of the crafts we serve — written in one country, mixed in another, mastered in a third, the label in a fourth, the whole thing held together by files crossing borders. The work already travels by link. We simply think the listening, the noting, the humming and the deciding should travel with it, in the same frame, at full fidelity, for as long as the song needs to become itself.

There is a thread running through all of this, though, that deserves its own essay. Every feature above invites more people, earlier, into work that is not finished — and unfinished work is precisely the work you guard. Sometimes a link, however unguessable, wants a guest list. That is the companion piece to this one: a guest list for a link.