You sing something you're happy with, play the recording back, and it's flat. Not everywhere — just enough that you can hear it, and now you can't unhear it. The performance felt fine while you were doing it. The recording says otherwise.
Almost everyone runs into this, and almost everyone draws the wrong conclusion from it: that they simply can't sing. The real explanation is much more boring, and much more fixable.
Why the recording sounds worse than the performance did
Three separate things are working against you, and only one of them is about your voice.
You have never actually heard yourself sing
While you sing, most of what reaches your ears travels through the bones of your skull rather than through the air. Bone conducts low frequencies far better than high ones, so your own voice arrives warmer, rounder and fuller than it really is. A recording removes that channel entirely and gives you only the air-conducted version — the one every listener has always heard.
This is why the first playback is so jarring, and it is also why it is so useful. You are not hearing a worse voice. You are hearing the actual one for the first time.
You had nothing to be in tune with
Pitch is relative. A note is only flat or sharp in relation to some reference, and if you practise unaccompanied, that reference is your own memory of where the note was a few seconds ago. Memory drifts. Over a phrase of eight or ten notes, a drift of ten cents per note compounds into something obviously wrong by the end, and you will not notice it happening because each individual step felt correct.
This is the entire reason a tanpura drone exists in Indian classical practice, and the reason Western singers practise against a piano. A constant reference note means every pitch you sing is measured against something fixed rather than against your own drift.
You had no feedback loop
Skills improve when the gap between doing something and finding out whether it worked is short. Sing a phrase, hear nothing back, sing it again — that is not practice, it is repetition, and repetition without feedback happily cements whatever you were already doing wrong.
What "in tune" actually means
Musicians measure pitch error in cents. A semitone — the distance between two adjacent piano keys — is 100 cents. Some rough thresholds worth knowing:
- Under 5 cents — inaudible. Nobody, including you, can hear this.
- 5 to 15 cents — professional territory. Studio vocals typically sit here.
- 15 to 30 cents — noticeable to a trained ear, fine for most singing.
- 30 to 50 cents — clearly out of tune to most listeners.
- Over 50 cents — you are closer to the neighbouring note than the one you meant.
That last threshold matters more than people expect. Past 50 cents you are not singing the note badly; you are singing a different note. It is also why "just sing louder and more confidently" is such poor advice — confidence does not move pitch.
A 15-minute daily routine
The structure below is the one used in Indian classical training, which has spent a very long time working out how to build pitch accuracy from scratch. It works just as well if you have never touched a raga in your life.
Minutes 1–3: warm up, quietly
Hum a five-note pattern up and back down — Sa Re Ga Ma Pa Ma Ga Re Sa, or simply do-re-mi-fa-sol and back. Lips closed, jaw loose, no volume. Move it up a semitone and repeat, four or five times.
The point is circulation and coordination, not sound. Singing loudly on a cold voice is how people end up hoarse by Thursday.
Minutes 4–9: alankars against a drone
This is the part that actually builds accuracy, and the part everyone skips.
An alankar (or palta) is a short melodic pattern moved systematically up and down a scale. The simplest one takes three notes at a time — Sa Re Ga, then Re Ga Ma, then Ga Ma Pa, and so on to the top of the octave, then the mirror image coming down. There are dozens of them, ranging from that to patterns that skip intervals and zig-zag.
Two rules make the difference between practice and noise:
- Keep a drone running. A tanpura, a keyboard holding the tonic, an app — anything constant. Without it you are measuring against your own memory again.
- Go slower than feels necessary. If you cannot land the note cleanly at 60 beats per minute, you certainly cannot at 100; you are just moving past the mistakes too quickly to hear them. Speed is a result of accuracy, never a route to it.
Minutes 10–13: one song, one phrase
Pick a single phrase — not a whole song — and work it. Simple traditional melodies are ideal here because the intervals are gentle and the shapes are predictable, which lets you concentrate on pitch rather than on remembering what comes next.
Minutes 14–15: listen back
Non-negotiable, and the step almost everyone drops. Record one pass of what you just practised and listen to it. Not to judge it — to locate the specific notes that drifted. Two minutes of honest listening will tell you more than another twenty minutes of singing.
The mistakes that hold people back longest
Practising in the wrong key. If your tonic sits too high, everything above the fifth becomes a strain and you will go flat trying to reach it. Too low and you lose support at the bottom. Find your comfortable range first and set your tonic so the exercise sits inside it.
Only ever going up. Ascending patterns are more satisfying, so people practise those and neglect descending ones. Descending phrases are where most singers go sharp, precisely because they get less attention.
Confusing volume with support. Getting louder to reach a high note almost always pushes it flat. The note needs breath underneath it, not force behind it.
Practising for two hours on Sunday. Fifteen focused minutes daily beats a weekend marathon comfortably. Pitch accuracy is motor learning, and motor learning consolidates between sessions rather than during them.
Where autotune fits — and where it doesn't
Pitch correction is a legitimate production tool. Essentially every commercial vocal you have heard in the last twenty years has had some applied, usually gently enough that you would never identify it. Used at full strength with an instant retune it stops being a correction and becomes an effect in its own right — the hard, stepped, robotic sound that has been a deliberate stylistic choice in pop and hip-hop for two decades.
What it cannot do is teach you anything. Correction moves the note after the fact; it does not tell you that you were 40 cents flat, and it certainly does not help you be less flat next time. Both uses are valid — just do not mistake one for the other. Correct the recording you are releasing. Practise the voice you are keeping.
Doing all of this in a browser tab
We built Voice Studio because the pieces above normally mean four separate things: a drone app, a metronome, a recorder, and something to tell you how far off you actually were.
It runs entirely on your own device — the recording never gets uploaded, because there is no upload. In it you get:
- A tanpura drone in any tonic, plus six talas if you want a rhythm cycle under your practice.
- Fourteen traditional alankars across twelve ragas, which generate automatically for whichever combination you pick, at whatever tempo you set.
- A live pitch graph showing the notes you are aiming for and the line you are actually singing, so the feedback gap collapses to nothing.
- Scoring in cents after each attempt, marking every note green, red or missed, so you know exactly which ones to go back to.
- A range finder that works out your comfortable tonic instead of making you guess.
- A dozen public-domain melodies in sargam notation to practise against — and a teleprompter for pasting in the words of anything else you want to sing.
And when you do want the recording to sound finished rather than honest, the full effect rack is there too: correction from subtle to fully robotic, harmonies that follow the scale, reverb, character voices, seven backing beats, and export to audio or a video you can actually post.
It's free, there is no account, and nothing you sing leaves your device.
