Voice Tone Test
Read naturally or hold a comfortable note. Keep the same microphone, distance and gain when comparing takes.
Microphone off0.0 s measuredUp to 20 MB / 120 seconds. Codecs depend on browser support. Choose a short solo voice sample; music, noise, effects and microphone processing can alter results. Audio stays local. Stereo channels are averaged; opposite phases can cancel.
No file selected.
No take A kept. Results remain in this tab only.
Input and analysis method
Fundamental detection50–1,500 Hz; Hann-windowed4,096-point spectrum. Spectral centroid is a magnitude-weighted frequency, not a measure of warmth or talent. Periodicity is the detector's normalized autocorrelation on detected frames, not clarity, intelligibility or clinical HNR. Levels are relative digital dBFS, not acoustic SPL. No recording or upload occurs during live measurement.
Held-note consistency is an app-defined0–10 practice score:60% × max(0,1−median absolute pitch deviation/100 cents),30% × voiced time fraction,10% × (1−clipping-frame fraction). Requires 3 voiced seconds; natural vibrato can lower it. It is not overall singing quality, attractiveness, vocal health or a professional rating. Speech has no quality score.
Ready when you are.
Related tools
- Voice Frequency Test — Measure and interpret a short sample’s fundamental-frequency distribution.
- AI Voice Analysis — Explain local acoustic analysis and the limits behind an existing AI-labeled page.
- Rate My Voice — Explain and use the optional 0–10 held-note consistency formula.
The Voice Tone Test provides a measured snapshot of a short voice sample. It reports detected fundamental pitch, spectral centroid, frequency-band energy proportions and digital level. Use it to compare controlled takes and understand what changed in the recorded signal, without turning those measurements into an attractiveness or personality judgment.
Choose a task before recording
- Select Natural speech or One comfortable held note.
- Choose a 10, 20 or 30-second microphone sample.
- Check the microphone, distance and input level.
- Start, complete the task and stop when finished.
- Read pitch, spectral and level results as separate measurements.
Natural speech is appropriate for a repeated phrase or short passage. Held-note mode is appropriate for one comfortable sustained sound. A conversation and a vowel are different tasks, so do not compare their results as though only the voice changed.
A supported local audio file is another input option. The file limit is 20 MB and 120 seconds, with mono or stereo audio and additional decoded-sample and frame caps. Browser support for the actual codec determines whether decoding succeeds.
Separate pitch from spectral balance
Fundamental frequency estimates the repeating rate of the detected voiced sound. Spectral centroid describes a magnitude-weighted center of the analyzed frequency content. These are different properties.
Two takes can have a similar fundamental while showing different centroids. Microphone distance, articulation, room reflections, processing and background sound can contribute to that difference.
A higher centroid can describe a shift toward higher-frequency content within this analysis, but it does not establish a universally brighter, healthier or more attractive voice. Perceived tone includes context that a single numerical feature cannot capture.
Read the measurement panel
| Measurement | What it describes |
|---|---|
| Median fundamental | The middle retained pitch estimate |
| 5th–95th detected pitch | A central span of detected pitches in this sample |
| Median absolute pitch deviation | Typical absolute cents distance from the sample’s median pitch |
| Spectral centroid | A weighted frequency center, not a voice-type label |
| Digital RMS and sample peak | Signal levels relative to digital full scale |
At least one detected voiced second is required for a pitch report. Pauses, unvoiced consonants and uncertain frames do not supply reliable fundamental estimates.
The mean periodicity value summarizes the retained detected frames. Because it depends on which frames passed detection, it is not an independent clinical voice-quality score.
Understand the three frequency bands
The displayed bands cover approximately 50–250 Hz, 250–2,000 Hz and 2,000–8,000 Hz, limited by the input’s available frequency range. The proportions describe how the analyzed spectral energy is distributed across those bands.
A larger low-band proportion does not prove a lower habitual speaking pitch. Band energy includes spectral components beyond the fundamental, and an altered input setup can change the balance.
Likewise, the high-band proportion is not a measure of clarity, confidence or intelligibility. Use it as a descriptive comparison under controlled conditions rather than assigning a personal trait to a percentage.
Compare take A with take B
After a useful result, choose Keep as take A. Make a second sample with the same task and recording setup. The comparison reports B minus A for median fundamental, spectral centroid and digital RMS.
If take A has a centroid of 1,200 Hz and take B has 1,450 Hz, the displayed difference is +250 Hz. That is an arithmetic change in this feature, not proof that B sounds better.
Keep microphone gain, distance, room and spoken material consistent. Changing the microphone or adding an effect can overwhelm the difference you wanted to inspect. A/B comparison also requires matching task modes.
Take A remains in this tab’s working memory. It is not a saved recording or a permanent account entry. Reset clears the working measurements and selected file.
Know what the tool does not judge
This page does not provide an attractive-voice test, a clinical diagnosis, a gender classification or an emotion detector. It uses local signal analysis rather than a trained model that understands speech or musical interpretation.
Held-note mode may show a custom consistency score after at least three voiced seconds. That score describes a limited exercise, not tone beauty. The Rate My Voice page explains the formula in detail.
For a simpler speaking-frequency summary, use the Voice Frequency Test. For comfortable singing endpoints, use the Vocal Range Test.
Export and privacy
Copy the measurements or download CSV and JSON. CSV provides scalar measurements; JSON also preserves the band values and fuller analysis structure. Neither export contains the original audio.
The tool processes microphone input and selected files locally. It does not upload a voice sample or create a public analysis page. Stop input and export anything you want to keep before resetting.
Frequently asked questions
Does this tell me whether my voice sounds good?
No. It describes measured signal features. Aesthetic preference and communication quality cannot be reduced to the centroid, fundamental or band proportions.
Are the dB values sound-pressure levels?
No. They are dBFS values relative to digital full scale. Without an acoustic calibration, they are not room sound levels or a hearing-safety measurement.
Why does moving closer to the microphone change the result?
Microphone response, gain, room sound and distance affect the recorded signal. Keep those conditions stable when comparing takes.
Can natural speech receive a quality score?
No. Speech mode reports measurements without a quality rating. A separate held-note consistency formula is available only with enough detected voiced input.
Why might microphone mode be unavailable?
It requires permission, a secure context and browser support for AudioWorklet. A supported local file can be an alternative, subject to the file and decoding limits.