How the test works

What we measure, how accurately, the norms behind each percentile, and what neither the test nor the training can do.

It isn't a medical test

It measures pitch, not the health of your voice. If you're hoarse for more than two weeks, speaking hurts, or you're an adult man whose voice never dropped, see an ENT doctor or a speech-language pathologist.

What we measure

Your median speaking fundamental frequency: how many times a second your vocal folds vibrate while you read aloud, in hertz, taken as the median over about 8 seconds of voiced speech (at least 3). The median is the pitch you spend most of your time around, so a few high or low words don't move it much. This is the standard clinical measure of how high or low a voice sits. We also show the 10th to 90th percentile of the readings, your range while reading, and the nearest musical note.

How it's measured

Nothing is recorded: each frame is analysed and thrown away. Only the resulting numbers are saved.

How accurate it is

We test the exact pipeline the browser runs against synthetic speech whose true pitch is known. The speech has syllables, pauses, intonation, jitter and vowel-like formants, at speaking pitches from 80 to 210 Hz. There are three conditions: clean, a phone-like microphone that weakens low frequencies, and a noisy room through a cheap phone microphone. In every case the measured median is within 2 Hz of the truth; on our tuning runs the largest error was 1.3 Hz. Real microphones and rooms vary more than any bench, so if a result looks odd, retake it somewhere quieter or with a headset.

The norms behind the percentiles

Adult speaking pitch from Leung, Oates, Papp and Chan (2022): 135 men and 244 women aged 18 to 60, speakers of Australian English. Men averaged 115 Hz (standard deviation 21 Hz) and women 199 Hz (standard deviation 28Hz). Pitch is spread more like a log-normal curve than a bell curve, so we use the log-normal with the same mean and standard deviation. “Deeper than N%” is the share of that curve above your pitch. For example, 100 Hz is deeper than 75% of adult men, and 130 Hz is deeper than 22%.

Bands: deeper than 90% or more is very deep; 65 to 89% deep; 35 to 64% average; 10 to 34% a bit higher; under 10% higher. These are our own cut-offs for describing a result, not categories from the research.

Limits. Average pitch differs between languages, regions and ages by several hertz, and speakers vary from day to day. Treat a percentile as a rough guide, never as a ranking of voices. The norms are for adults: teenage voices keep deepening until about 18.

What training can and can't do

Your vocal folds' length and mass were set by puberty and hormones; no exercise changes them. What practice changes is where in your range you habitually speak, and how full your voice sounds. The best evidence comes from voice training for trans men:

We found no study of pitch-lowering practice in cisgender men. So we promise modest change, honestly measured. The program aims a little below your check-in pitch, never closer than two semitones to your floor (the lowest easy note you can sigh down to), and moves its aim a quarter of a semitone at a time.

Sources

Questions about the method? Everything the measurement does is described above; nothing about it is secret. Back to the test.