Our Testing Methodology

This page describes how the tools on VoiceRangeTest.com are checked, and what the results show. Why a browser reading can still differ from the truth on your own device is explained in accuracy and limitations. The tests are designed and run by Harlow, who runs the site.

How a tool is tested

Every test run starts from a clean WordPress install in WordPress Playground, with the tool plugin installed as it would be on the live site. A script then drives a headless Microsoft Edge browser. Instead of a real microphone, Edge gets a fake one that plays a prepared audio file, so each test knows exactly which note, frequency or noise the tool is hearing.

Each check has an expected result and records the actual result next to it. It is marked PASS only when the actual output matches within a stated tolerance, for example ±0.5 Hz for a 220 Hz tone, ±3 cents for a tone 20 cents sharp, or ±3% for average speaking pitch. Anything that could not be tried is marked NOT TESTED rather than assumed to work. A version is released only with 0 failures.

Besides accuracy, the suites check the parts people actually touch: error messages for empty or reversed notes, what happens when microphone permission is refused, that silence and noise never produce a fake note, that two copies of a tool on one page don’t fight over the microphone, that the page has no script errors, and that the layout fits phone screens 360 to 390 pixels wide without sideways scrolling.

Reference signals

• Steady tones at exact frequencies: A3 (220 Hz), a bass E1 (41.2 Hz) and a high C7 (2,093 Hz).
• Detuned tones: A4 20 cents sharp (445.1 Hz), A3 15 cents flat and 35 cents sharp.
• Synthetic singing with five harmonics and vibrato, including a 16-note C4–C5 melody and a slow glide from A2 to E4.
• Synthetic speech with a known average pitch of 119.4 Hz and 208.9 Hz.
• Hard cases: a very quiet note, ±40-cent vibrato, silence, white noise, a non-audio file and a blocked microphone.

The pitch engine

Six tools share one pitch engine: the Voice Range Test, Pitch Detector, Note Identifier, Pitch Accuracy Test, Voice Type Test and Voice Hz Test. Sound reaches it through the browser’s Web Audio API, and it works like this:

• It uses the YIN method (threshold 0.15), which is less prone to octave jumps than the plain autocorrelation used by the older versions.
• A reading counts only if its clarity is at least 0.8 and the sound is above a quiet-signal gate; a median of three readings smooths out single errors.
• The range tools keep a note only after it has been held steadily for about 0.8 seconds, so slides and cracks don’t become your lowest or highest note. The Pitch Detector lists a held note after 0.3 seconds.
• Notes are named with A4 = 440 Hz and equal temperament (each semitone is a ratio of 2^1/12). The Pitch Detector lets you move A4 between 415 and 466 Hz.
• The tools ask the browser to switch off echo cancellation, which can bend a sung note.

Tool-by-tool results

Accuracy results

• Pitch Detector: 220 Hz read as 220.0 Hz (0 cents); 445.1 Hz read as +20 cents; E1 read as 41.2 Hz; C7 read as 2,094.7 Hz (+1 cent). A very quiet A3 was found on High sensitivity and ignored on Low, as intended.
• Voice Hz Test: synthetic speech at 119.4 Hz measured 119 Hz (0.3% error) and 208.9 Hz measured 210 Hz (0.5% error).
• Note Identifier: all 16 melody notes found in the right order from an audio file.
• Pitch Accuracy Test: a note 15 cents flat was measured at −15 cents and one 35 cents sharp at +35 cents (33 points). A note with ±40-cent vibrato was scored at −5 cents, close to its true centre. In the hearing test a scripted listener who can hear 12 cents came out at 7.1 cents, inside the accepted window of 5 to 25 cents: the hearing result is a rough estimate.
• Voice Range Test and Voice Type Test: the A2→E4 glide gave A2–E4, and a range sitting equally between two types is reported as “between baritone and bass”.

Testing also catches real bugs before release. The Voice Type Test once named a single voice type while its own chart showed a tie; the Pitch Accuracy Test’s level bar never moved because of a shared code hook; the Note Identifier named one “most frequent note” when two tied; and the chart printed 1,046.50 Hz in one place and 1046.50 Hz in another. All were fixed and re-tested.

Tools not yet in the documented tests

These older tools were built before the current process and have no test report yet, so this page gives no accuracy figures for them:

• Microphone tools: Deep Voice Test, Frequency Finder, Vocal Scale Finder and the microphone mode of the Vocal Range Calculator. They use their own, simpler pitch methods.
• Audio-file tool: the Song Key Finder.
• Playback and practice tools: Tone Generator, Audio Frequency Test, Online Metronome, Interval Ear Training and Vocal Warm-Up Generator. Metronome timing has not been measured.

What was checked for all of them, on 27 September 2026, is their live code: none contains network or recording code, and the Song Key Finder reads your file inside the browser. They will get the same test suite when they are rebuilt. Until then, treat their readings as rough.

What testing does not cover

All tests so far ran in Microsoft Edge on Windows. Safari, Firefox, a physical Android phone, real voices have not been formally tested for any tool. A synthetic voice is cleaner than a real one: breathiness, vocal fry, room echo, a phone’s own noise suppression and cheap microphones can all make real results worse than the numbers above. That is why the site calls its results estimates, and why a voice type from a range is only a starting point.

How singer ranges are researched

Singer range articles follow the editorial guidelines: recordings and credible references are compared, and three things are kept apart: the range a singer uses regularly, notes that are documented but rare, and claims that cannot be verified. Where sources disagree, the article says so. The Singer Range Widget copies its numbers from each article’s own opening sentence or facts table, shows the documented range as a solid bar and wider reported claims as a dashed bar, and is checked against the article whenever the text changes.

Privacy: what was verified

Your browser asks before any page can use the microphone (getUserMedia always needs permission). On 27 September 2026 every live tool script was read: none contains code that sends data (fetch, XMLHttpRequest, sendBeacon or WebSocket) and none uses MediaRecorder, the browser’s recording feature. Audio is analysed in your browser and never recorded or uploaded. The Vocal Range Calculator keeps your last five results in local storage; four practice tools keep a short history until you close the tab. Details are on the data security page.

Retesting and corrections

• Every new version of a tested tool runs the full suite again before release, and new checks are added for each bug found.
• If a tool changes how it measures, this page and the tool’s own page are updated in the same step.
• A reader report is reproduced first, where possible with a test signal that shows the same problem.

If a tool gives a result that looks wrong, email contact@voicerangetest.com or use the contact page. Please include the tool, your device and browser, the microphone you used, and what you expected versus what happened. Confirmed errors are fixed and noted on the page concerned.

Last reviewed: 27 September 2026