Vocal Spectrum — Vocal Frequency Analyzer

Vocal Spectrum is a real-time vocal frequency analyzer that helps you see where your voice energy sits across the spectrum while you sing.

What You Can Read at a Glance

See low/mid/high energy balance instantly, check whether harmonics are stable, and spot where sibilants or breath noise spike in higher bands.

Why It Is Useful in Practice

Repeat the same phrase and compare your spectrum pattern across takes to hear and see improvements in tone consistency, transitions, and articulation.

Low—High
Graph
Spectrum floor49%
PEAK—LEVEL—CENTROID—RMS—

Use the microphone,
or upload an audio file.

How to Use Vocal Spectrum (for Vocal Practice)

Use it while singing to monitor frequency-energy distribution in real time, then replay and compare sections to verify what changed and what stayed consistent.

Live Feedback You Can Act On

  • Higher bars mean stronger energy in that frequency band, so you can quickly see where your tone is focused.
  • Peak-hold lines keep short transients visible, helping you catch quick consonant attacks and sharp bursts.
  • Log-frequency spacing keeps important low and mid vocal areas readable instead of visually compressed.

How to Read This Graph

  • X-axis (Hz, log): 80 Hz to 16 kHz. Guide lines at 125/250/500/1k/2k/4k/8k Hz help you anchor where energy clusters.
  • Y-axis (dBFS): 0 at the top; the bottom follows Sensitivity (about -30 dB at 0% to about -75 dB at 100%). Closer to 0 means stronger energy in that band for that frame.
  • Bars are FFT-bin aggregates: they show band energy, not note names and not a direct pitch detector output.

Panel Controls (Low/High, Graph, Sensitivity, Stats)

  • Low-High sets the visible analysis band (about 20 Hz to 22 kHz). Narrow the band to focus on one problem area, such as low resonance or upper sibilance.
  • Graph mode gives three views of the same data: Bar for fast band-by-band reading, Line for smooth contour tracking, and Circle for radial pattern comparison across takes.
  • Spectrum floor (0–100%, independent of Pitch gate) controls the display floor in dB. Higher values reveal weaker partials and breath components; lower values clean up low-level noise so major energy trends are easier to see.
  • The stats chips summarize the current frame inside the selected band: PEAK (dominant frequency), LEVEL (peak dB), CENTROID (amplitude-weighted mean frequency), and RMS (root mean square of retained bin magnitudes). Both the band and floor affect these values; they do not measure calibrated loudness.
  • Panel choices are saved in browser local storage, so your preferred range, graph type, and sensitivity are restored when you reopen the tool.

What Changes When You Adjust the Panel

  • Changing Low-High can shift PEAK and CENTROID immediately because bins outside the selected band are excluded from stats and drawing.
  • Sensitivity mainly affects weak content visibility and readout inclusion threshold, which is useful when comparing quiet breathy takes versus stronger projected takes.
  • Peak-hold behavior stays consistent across Bar, Line, and Circle views: peaks stay visible briefly, then decay gradually so transients remain readable.

Playback Insights Across the Full Take

  • Recorded spectrum history is synced to playback, so you can inspect any point in the take (up to 10 minutes).
  • This is ideal for A/B checks: before vs. after technique changes, mic position changes, or warm-up effects.

Useful Vocal Signals to Watch

  • Harmonic structure: check whether overtone peaks remain stable as you sustain notes.
  • Register transitions: identify where chest-to-head shifts cause abrupt spectrum changes.
  • Sibilance control: monitor high-frequency peaks during 'S', 'Z', and 'CH' sounds.
  • Breath noise: if 10 kHz+ stays elevated, adjust mic angle, distance, or pop filter.

Reference: Typical Fundamental (F0) Ranges by Voice Type

  • Bass: E2-E4 (about 82-330 Hz), Baritone: G2-G4 (about 98-392 Hz), Tenor: C3-C5 (about 131-523 Hz).
  • Alto/Contralto: F3-F5 (about 175-698 Hz), Mezzo-soprano: A3-A5 (about 220-880 Hz), Soprano: C4-C6 (about 262-1047 Hz).
  • Use this as a practical orientation only. Pop/musical/CCM usage can differ, and individuals often overlap multiple categories.
  • When you sing a sustained pitch, the lowest clear strong peak is often near the fundamental. If it is far from your intended note area, re-check pitch target and support.

Quick Practice Workflow

  • Sing one phrase at medium volume, then louder, and compare whether harmonics rise smoothly or become overly spiky.
  • Repeat a difficult transition note and watch where the spectrum shape jumps; re-test after vowel or resonance adjustments.
  • Record two takes with different mic distance, then scrub playback to compare sibilance and breath-noise peaks section by section.

Tools That Pair Well

  • Use Vocal Pitch Monitor for pitch accuracy and Vocal Spectrum for tone-shaping and energy balance.
  • Use Singing Recorder to capture takes, then review section by section to build repeatable vocal control.

What the chips are counting

The graph is an FFT of the current window. Each bar is one bin, in dB, on a log frequency axis from the Low–High you set. Peak-hold remembers a bin's recent high and lets it fall, so a consonant does not vanish in a single frame.

ChipOn this page
dBFSLevel relative to digital full scale. 0 is the top of the axis. Sung audio sits below it.
PEAKThe strongest bin inside Low–High, after the sensitivity floor. On a bright vowel it can be a harmonic, not the note you think you are singing.
CENTROIDThe linear-amplitude-weighted mean frequency of bins inside the selected band above the floor. It is not a measured formant or vowel name.
RMSThe root mean square of magnitudes in the selected band above the floor. It is not waveform RMS, SPL, or LUFS and changes with the settings.

Hold one pitch and switch ee, ah, and oo. Changes in CENTROID and upper bars can help compare those vowels, but the display does not identify vowels or estimate formants; pitch, microphone response, and floor settings also affect it. Phone microphones often drop energy below a few hundred hertz, so a thin low end on a loud low note is sometimes the mic, not the vowel.

Questions

Does Vocal Spectrum upload my microphone audio?

No. The bars are computed in the browser from the microphone, or from a take already loaded in the Controller.

Why does Sensitivity change the picture and not the sound?

Sensitivity sets a dB floor. Bins under that floor are left out of the drawing and out of PEAK and CENTROID. The audio you hear is unchanged. Low–High does the same kind of cut at the edges of the band, so both controls can move the chips without you singing any differently.

Why is there one tall spike and almost nothing else?

A single narrow spike, especially near 50 or 60 Hz and its multiples, is often hum. A sung note usually shows a fundamental and a row of harmonics getting weaker as they go up. If the spike sits on a harmonic of a bright vowel, PEAK may be that harmonic. The note name is on the pitch monitor, not on this graph.

Can I see formants or vibrato here?

You can see a vowel change as a shift in which upper bins are strong, and CENTROID will move with it. You cannot read a textbook F1/F2 off these bars. Those published numbers are mostly male speech averages and will not match every singer. Vibrato is not a band getting wider. This view is one spectrum at a time, so a wobble makes PEAK step between neighboring bins. The pitch monitor draws that wobble across time.

Can I compare my spectrum with another singer's recording?

Only if that recording is a take you are allowed to upload. Playback then drives this graph through that file. There is no second trace and no way to pin two singers on screen together. For your own before-and-after, record two takes and scrub each one.

The bars never move — did I deny the microphone?

Most likely, or the prompt was dismissed without a choice. Check the permission icon in the address bar, set this site's microphone access to allow, and reload. A take already loaded in the Controller will still drive the graph during playback either way.