Get an AI voice analysis of your singing or speaking in about 15 seconds. Sing a few long notes or read a short passage, and this free AI voice analyzer shows how many notes were in tune, how steady they were, your range, pace and loudness, with a plain summary. It runs as audio processing in your browser.
AI Voice Analyzer
Sing a few long notes or read a short passage for about 15 seconds. You get measured numbers and a plain-language summary. The analysis runs as audio signal processing in your browser.
- Choose
- Record
- Result
Step 1 Choose what to analyse
Then sing
Sing 3 to 6 long notes, or a slow phrase, on "ah" or "la". Hold each note for at least half a second.
You get: how many held notes were in tune (within 25 cents), steadiness, vibrato, drift, the range you used and loudness.
Then read this aloud in your normal voice
Every Saturday I take the early train to the coast. I buy a coffee at the station, find a seat by the window and watch the fields turn into small towns. By nine o'clock the sea appears, grey at first and then bright blue.
You get: median pitch, pitch span and variation, pace, pauses and loudness.
Best results: a quiet room, your normal volume, the same distance from the microphone.
Step 2 Record your voice
Time left
Analysed in this browser with audio signal processing. Nothing is uploaded or saved.
Choose Singing or Speaking in Step 1, then press Start recording.
Keyboard: Space or Enter on the button starts and stops. Esc stops a recording.
Your analysis
No analysis yet
Finish Step 2 to see your numbers, 3 to 5 observations about your take and how reliable the recording was.
What the analyser found
Held notes (offset from the nearest note)
Recording quality
Compared with your previous take
Kept only while this page is open. Compare takes made in the same room at the same distance.
How this works: the analyser measures your take with audio signal processing in this browser (YIN pitch detection and loudness levels) and writes the summary from fixed text rules. It uses no machine-learning model and uploads nothing. It does not rate talent, attractiveness, voice type or health, and it is not a medical test. Thresholds such as 25 cents for "in tune" are this site's conventions.
Related Tools
What This AI Voice Analyzer Measures
The analyzer measures pitch and loudness in your recording and turns the numbers into short sentences. The analysis runs as audio processing in your browser. It finds your pitch about 25 times a second with the YIN method, measures loudness, and fixed text rules write the summary. It uses no machine-learning model and sends no audio anywhere.
| Measurement | What it tells you | How it is worked out |
|---|---|---|
| Notes in tune | How many held notes hit a note | Each note held for 0.3 seconds or longer is compared with the nearest note of standard tuning (A4 = 440 Hz). Within 25 cents counts as in tune. |
| Steadiness | How much a held note wavers | The spread of the pitch in cents, smoothed over 0.2 seconds, after any slow slide is removed. |
| Vibrato | A regular pulse in the pitch | 4 to 8 cycles per second, at least 15 cents either side. |
| Drift | Whether a note slides up or down | The straight-line trend across a note of 0.8 seconds or longer. |
| Median pitch and span | Your typical speaking pitch and how much it moves | The middle pitch reading in Hz, and the 10th to 90th percentile in semitones. |
| Pace and pauses | How fast you speak and how often you stop | Loudness peaks per second of speaking time, and silences of 0.25 seconds or longer. |
| Loudness range | How much your volume changes | The 10th to 90th percentile of the level, in decibels. |
A cent is one hundredth of a semitone. The thresholds, such as 25 cents for in tune, are this site’s conventions, and the result card always shows the exact numbers behind each sentence.
Can AI Rate My Voice or Singing?
This AI voice rater measures your take; it does not judge your talent. You get no single score out of 10 or 100. Instead you see facts you can check and improve, such as “4 of 5 held notes in tune” or “steadiness 8 cents”. The analyzer gives no verdict on talent, attractiveness, voice type or health.
Each result also shows a recording quality of High, Medium or Low. It rates the recording, not your voice: seconds of clear voice, background noise and microphone overload. With Low quality, treat the numbers as rough.
How to Read Your Singing Result
The big number, for example 4/5, is how many of your held notes were within 25 cents of the nearest note. Below it, every held note is listed with its name, its offset and a small bar. A marker left of centre means flat, right of centre means sharp, and the green zone is in tune.
- Offset: “12 cents sharp” means the middle of that note sat 12 cents above the nearest note. The furthest note is named in the first observation.
- Steadiness: under 10 cents is very steady, 10 to 15 cents is steady, and above 15 cents is wavering on this tool’s scale.
- Vibrato: a regular pulse is reported on its own, with its speed and width, so it does not count as wavering.
- Drift: a long note that slides up or down is named in the observations.
- Range used: your lowest and highest held notes in this take, not your full vocal range.
Offsets are measured against standard tuning. If you sing without accompaniment, your whole take can sit a little above or below standard tuning and still sound in tune with itself. Singing along to a track tuned to A4 = 440 Hz gives the fairest note accuracy.
How to Read Your Speaking Result
In Speaking mode the big number is your median speaking pitch in Hz, with the nearest note beside it. The median is the middle of all accepted pitch readings, so one laugh or stressed word barely moves it.
- Pitch span: how far your pitch moves while you talk, in semitones. A wider span means more rise and fall in your voice.
- Pace: syllable beats per second of speaking time, counted from loudness peaks. It is an estimate, so compare your own takes rather than chasing a target.
- Pauses: the number of silences of 0.25 seconds or longer, and the longest one.
How to Get an Accurate Analysis
- Stay quiet for the first second. The analyzer measures the room while the level meter says Measuring room. Start when the status says Recording.
- Use a quiet room. Fans, music and other voices can hide your notes.
- Hold your notes. Sing each note for at least half a second on one vowel, such as “ah”. Fast runs and slides are not counted as held notes.
- Keep a steady distance. Use the same microphone at the same distance every time. If the meter shows Too loud, move back a little.
- Record again and compare. From the second take in the same mode, the result shows the change, for example “Notes in tune: 3/5 → 4/5”. Use Copy result to keep a record.
Describe My Voice: What the Summary Says
The summary describes your take in measured terms only. It says which notes you held, how steady they were, how wide your pitch moved and how loud you were. It does not describe tone colour, timbre or resonance, because the analyzer does not measure them, and it never calls a voice good, bad or attractive.
Frequently Asked Questions
Is this AI voice analysis free?
Yes. The analyzer is free to use and needs no account or app. It runs in an up-to-date browser on a device with a microphone. Allow microphone access when your browser asks, and the analysis starts after a one-second room check.
Does the AI voice analyzer use machine learning?
No. It uses audio signal processing: the YIN pitch method and loudness measurement, followed by fixed text rules. The rules are fixed, so the same numbers always produce the same sentences. Nothing is learned from other people’s voices, and no online AI service is called.
Is my recording uploaded or saved?
No recording is made. The analyzer reads the live microphone signal in your browser and keeps only numbers while the page is open. It creates no audio file, and closing or reloading the page clears everything, including the comparison with your previous take.
Can it rate my voice out of 10?
No. A single score would hide what to work on, so the analyzer shows separate measurements instead: notes in tune, offset, steadiness, vibrato, range and loudness for singing, and pitch, span, pace and pauses for speaking. Each one comes with its exact number.
How accurate is the note accuracy?
In this site’s tests with synthetic sung notes played into Chrome and Edge, measured offsets came within 5 cents of the true value. Real voices, microphones and rooms add error, and the tool has not been benchmarked on recorded human singing, so no accuracy percentage is given.
Can it tell my voice type or full vocal range?
No. It reports the range you used in one take and names no voice type. To find your lowest and highest notes on purpose, use the Vocal Range Test linked above.
Why did it find no held notes?
A held note needs at least 0.3 seconds of steady pitch. Fast runs, slides, whispering and speaking in Singing mode break notes into pieces that are too short. Sing slower, hold each note for half a second on “ah”, and try again.
Method and Author
How the analysis works. The analyzer estimates pitch with the YIN algorithm (de Cheveigné and Kawahara, 2002) about 25 times a second and refines each reading on the full-rate signal. Held notes, offsets, steadiness, vibrato, drift, pauses and loudness come from fixed formulas, and the sentences come from fixed templates. Note names use A4 at 440 Hz.
Conventions. In tune means within 25 cents of the nearest note. Steadiness labels are under 10 cents, 10 to 15 cents and above 15 cents. These are this site’s thresholds, not published standards.
Author. Written by John for Vocal Range Tester. Last updated: 30 September 2026.