May 2, 2026 · 5 min read
Voice Notes Are Songs Before They Know It
On capturing the thing while it is still shapeless, and why editing too early kills it.
The best line in a song is very often the first way you said it, before you knew it was going in a song.
This is not romanticism about spontaneity. It is a practical observation about what happens to language when you become aware it is being written down. Spoken, you say: nobody's said anything about it and I haven't asked. Written, an hour later, you say: we keep our silence like a debt. One of those sounds like a person. The other sounds like someone trying.
Capture before you shape
The purpose of a voice note is not to record a song. It is to get the thing out of your head at the speed of speech, before the editor wakes up.
A few things that help:
Talk to somebody. Not to a microphone. Pick a real person and tell them what happened. Your voice does something different when it has a listener in mind, even an imagined one.
Do not start over. The instinct to restart the take is the editor arriving early. False starts are fine. Repetition is fine. The repetition is often the chorus telling you what it is.
Say the boring parts. The make of the car. The time on the clock. What the room smelled like. These are the details you will not invent later, and they are worth more than anything you will invent later.
Stop before it is finished. Ending on an unresolved thought gives you a reason to come back.
The transcript is not the song
Here is where people go wrong in the other direction: they transcribe the note and treat the transcript as a draft. It is not. It is evidence. A transcript of natural speech is repetitive, tenseless, and structurally shapeless, and it should be — that is what speech is.
What you are looking for in it is three or four moments, not a shape:
Everything else is scaffolding. Take the four moments and build.
Why we show you the transcript first
Coroas transcribes a voice note and then stops, and shows you the text, and lets you edit it before anything is generated. That pause is deliberate and slightly inconvenient on purpose.
Transcription is imperfect, and the words it gets wrong are usually the words that matter — proper nouns, place names, the mumbled thing at the end. More importantly, seeing your own speech as text is the first useful edit. You will cut a third of it yourself, immediately, and you will be right to.
On keeping the audio
We do not keep guest recordings. The audio is transcribed and discarded — nothing written to disk, nothing kept for later. If you have an account, we still do not keep it unless you explicitly ask us to save the original.
That is partly a privacy position and partly a songwriting one. A voice note is not a demo. It is the sound of you working something out, and most people write more freely knowing it is not being filed somewhere.
Start a song