It stops talking when it isn’t sure.
Strand is a guitar tuner for iPhone. When it can’t hear a string clearly it shows you nothing at all, because a guess is how people end up tightening a string until it snaps.
Not released yet. It will be free, with no account.
When it can hear the string, you get a word first.
Tighten or loosen, in plain type, before any number. Someone who picked up a guitar last week may not know which way an arrow points.
The number comes second, in cents. A cent is a hundredth of a semitone, so twelve cents flat is a small turn of the peg. The scale is drawn as a machinist’s steel rule.
There’s one kind of reading it never shows. The mistake that breaks strings is a tuner hearing your low E as the E an octave below: it tells you the string is far too flat, and you keep tightening. If Strand can’t tell which E it’s hearing, it shows nothing until it can.
In tune, the whole screen turns green.
You can see it from across the room, which helps when the phone is on a music stand and both your hands are on the guitar.
It goes green once the reading has held within 3 cents, and stays green until it drifts past 6. My first version lit up only inside a 1.5-cent band. On a string nobody was touching, that came to about 4% of the time, in flashes a tenth of a second long, and it never once felt in tune. With two thresholds it holds for about three quarters of a settled string.
Or read it as a strobe.
A strobe tuner shows how far off you are as movement rather than as a needle’s position, and some people tune better with one, so there’s a second design. The marker on the arc gives you the quick reading. The rings inside it follow the slower, more precise measurement: they turn while you’re off and stand still when you’re on.
Underneath it’s the same measurement, with the same rules about when to show nothing. You switch between the two in Settings, under Design.
You can tune it without looking.
Turn on Audio Tuning Mode and the reading comes out of the speaker instead: spoken readings, a two-tone beacon, drones and a metronome. Every control and every reading is labelled for VoiceOver.
Both are in the first version on purpose. In App Store reviews, blind guitarists have said the best-known tuner isn’t labelled, and that it’s why they won’t pay for it.
On screen, state is always shown by position, colour and a word together, so nothing depends on seeing green.
What it won’t do.
Each of these is a hard rule in the spec the app is built against, and each answers a complaint people have made about a tuner they’d already paid for. There are eight, numbered as they are in the spec.
- C1
Nothing appears before or over the tuner.
No ads or promotions, and it never asks you to rate it. It opens to the tuner.
- C2
It’s listening before you see it.
The microphone is running before the first frame is drawn. The spec allows 400 ms from launch.
- C3
One screen, doing one thing.
The extra tools, like drones, a metronome and an intonation check, open over the tuner, and you can hide them for good.
- C4
No number unless it’s sure.
Below its confidence threshold you get a status in words instead of a figure.
- C5
No “too noisy” message.
In a loud room the readings get slower and more averaged. It won’t tell you to find somewhere quieter.
- C6
It works without the screen.
Audio Tuning Mode and full VoiceOver labels are in the first version, as described above.
- C7
Free, with nothing to unlock.
There’s no in-app purchase code in it, and no paid tier for a feature to move into.
- C8
No network, no analytics.
It never connects to anything. Airplane mode changes nothing.
The numbers.
I test Strand against recordings where the right answer is already known, and the build fails if the error rate goes over its limit. These are the current results. The app shows the same four under About.
Measured on a Mac, from my recordings and test tones. Not yet on a phone.
Trust
0of 202
A lock is the moment it declares a string in tune. This counts the locks that named the wrong note or had you turning the wrong way.
Target: zero wrong. No percentage claimed yet; see below.
Speed
87ms
From plucking to the first reading, for a typical note. 87 ms is the length of the stretch of sound the analysis needs, so a reading can’t arrive sooner. The slowest 5% take up to 258 ms.
26 of 216 single notes got no reading at all.
Targets: 100 ms typical, 300 ms slowest. The slowest-note target was 100 ms until I saw that meeting it meant reading notes from the pick attack, before the pitch settles.
Accuracy
0.09¢
How far off 95% of readings are, checked against an independent reference on recorded strings.
This is the fine reading, which appears only once a note has settled: about a quarter of the time a string is ringing.
Target: within 0.1 ¢.
Steadiness
0.0003¢
How much the number wobbles when the note itself isn’t changing at all.
Measured on a synthetic tone. A real string drifts as it rings, and the reading follows it.
Target: under 0.05 ¢.
Why the first figure is a count rather than a percentage: zero wrong out of 202 is still consistent with a real error rate of about 1.5%. Showing it’s under 0.1% would take roughly 3,000 locks, so until I have them, the count is all I’ll claim.
With several strings ringing at once, 27 of 36 notes got no reading at all. It can’t pull one string out of a chord, so it shows nothing.