Find your vocal range
Sing your lowest and highest note, or drop a recording.
Checking your browser…
Choose how to measure
-
1
Your lowest comfortable note
Slide down on an "ah" until your voice starts to crackle, come back up one note, and hold it for a second.
Held: —
-
2
Your highest comfortable note
Now go up. Stop where it still sounds like singing, not shouting or squeaking, and hold it.
Held: —
Your range
Voice types overlap and teachers draw the lines differently. Where your voice feels easy matters as much as the extremes, so treat this as a starting point, not a label.
Note list
| Start | Note | Length | Off by |
|---|
Getting a reading you can trust
- Warm up first. Two minutes of humming and lip trills adds a note or two at both ends compared with singing cold.
- Comfortable, not extreme. The note you can only reach with a crack or a whisper is not part of the range you would sing a song in.
- Hold it for a second. A note only counts when it stays within ±60 cents for a full second, so a slide past a note is not recorded as your range.
- About an arm's length from the mic. Too close and loud notes clip; too far and the room echo blurs the pitch. Turn music and fans off.
Approximate voice type ranges
These are the ranges most choir directors use when they sort new singers. Trained singers often go several notes past them.
| Voice type | Typical range | Where it tends to sit |
|---|---|---|
| Bass | E2 – E4 | Low, dark; the bottom line of a choir |
| Baritone | G2 – G4 | The most common adult male voice |
| Tenor | C3 – C5 | Bright male voice; most pop lead lines |
| Alto | F3 – F5 | Lower female voice, warm middle |
| Mezzo-soprano | A3 – A5 | The most common adult female voice |
| Soprano | C4 – C6 | High female voice; carries the melody |
How the pitch model compares with YIN
Most free range testers use YIN or autocorrelation, fast formulas that look for repetition in the waveform. This page runs CREPE, a neural network trained on labelled pitch. We measured both on synthetic sung melodies where the true pitch is known:
| Recording | CREPE full | CREPE tiny | YIN | pYIN |
|---|---|---|---|---|
| Bass melody, clean | 8.1¢ · 0.3% | 8.3¢ · 1.4% | 13.3¢ · 4.2% | 13.1¢ · 4.4% |
| Bass melody, breathy (10 dB SNR) | 8.4¢ · 0.9% | 8.6¢ · 1.9% | 13.5¢ · 4.0% | 13.5¢ · 4.3% |
| Soprano melody, clean | 7.7¢ · 1.0% | 7.9¢ · 2.2% | 8.6¢ · 2.7% | 9.0¢ · 2.5% |
| Soprano melody, very breathy (3 dB SNR) | 9.4¢ · 33% | 8.6¢ · 7.2% | 40.9¢ · 44% | 15.6¢ · 70% |
| Soprano melody over piano chords | 8.3¢ · 25% | 11.0¢ · 28% | 13.9¢ · 16% | 13.0¢ · 14% |
Each cell: median error on the notes it did track · share of sung frames it got wrong by more than 50 cents or did not track at all.
What this means for you: on a clean or breathy solo voice, CREPE lands closer to the true note and avoids the octave jumps YIN makes on soft high notes. Over a piano it is not magic: it declines more frames instead of guessing, which is why the curve has gaps. For songs with a backing track, strip the accompaniment first.
Questions
Is my voice recorded or uploaded?
No. The microphone audio goes straight from the browser's audio engine into the pitch model running in this tab, 10 frames every 100 ms; only the last second of sound is kept in memory, and only the pitch readings are kept after that. Nothing is saved to disk or sent to a server — the only downloads are the page and the model file. A dropped recording is decoded in memory the same way.
How accurate is it, and how is it different from other range testers?
On our 30-second test sweep from A2 to A5 the median error is 2.6 cents with the tiny model and 2.8 cents with the full model, and after smoothing only 1 frame in about 2,800 lands in the wrong octave. A cent is a hundredth of a semitone, and most listeners cannot hear a difference under about 10. Other free testers mostly use YIN or autocorrelation; the table above shows where the neural model helps (breathy and high notes) and where it does not (over accompaniment).
Can I test with music playing, or in a choir?
The model follows one pitch at a time. Backing music or other voices lower its confidence, and those stretches show as a red strip under the curve instead of wrong notes. For a sung track, run it through a vocal remover first and analyze the isolated vocal. For the microphone test, turn music off and use headphones if you want a reference note.
How is my voice type decided?
Your lowest and highest held notes are compared with six textbook ranges (bass E2–E4, baritone G2–G4, tenor C3–C5, alto F3–F5, mezzo-soprano A3–A5, soprano C4–C6). Each type is scored by how much of your range it covers, minus how far you fall outside it, with the low end counting more because it changes least with training. It is a starting point: teachers also listen to where your voice sounds easy and how it changes timbre.
Does it work for guitar, violin, flute or whistling?
Yes for anything that plays one note at a time: a single guitar string, violin, flute, trumpet, whistling or humming all give a clean curve. The model covers roughly 32 Hz to 1,975 Hz (C1 to B6), so very high whistles and piccolo can fall off the top. Strummed chords and piano chords are not supported — the model will pick one note or report low confidence.
Why does it download 89 MB the first time, and what if I have no WebGPU?
With WebGPU (recent Chrome, Edge or Safari on a computer), recordings are analyzed by CREPE full, an 89 MB network that takes about 3 s per 30 s of audio; it is downloaded once and then served from the browser cache. The live microphone dial always uses CREPE tiny (2 MB). Without WebGPU everything runs on the CPU with the tiny model, on one WebAssembly thread: around 18 s per 30 s of audio, so a 3-minute file needs about two minutes.