Vocal Pitch Monitor — Real-Time Intonation for Singers

A vocal pitch monitor in your tab: your fundamental is drawn in real time so you can work on singing in tune, vibrato, and stability — the same local WebAssembly/Web Audio analysis as the rest of the site, with no server-side processing of your voice. Save MIDI range, overlays, and sensitivity per browser; the canvas is HiDPI-aware. Use Fullscreen in the plot bar to focus on the trace (header + graph, sidebar hidden; Esc to exit). Ideal for warmups, phrase fixes, and passaggio checks without installing an app.

See exactly where your voice goes flat or sharp

The trace shows pitch drift through a line or phrase so you can spot sharp entrances, flat releases, and unstable vowels. Optional highlight overlays emphasize the nearest pitch row and a horizontal reference for the current tone — when Highlight is on, a brief drop in level can keep the last stable pitch visible for a few seconds before it fades, so short gaps do not feel visually noisy.

How singers use the pitch graph for intonation practice

Sing scales, arpeggios, or song fragments while watching the graph: aim for a smooth line on sustained notes and consistent peaks on repeated patterns.

Warm up your voice and track pitch stability over time

Use short sessions before rehearsals or gigs: lip trills, sirens, and long tones — the display helps you notice when your warmup has “settled” versus when you are still straining.

Works for all voice types — soprano, alto, tenor, bass

The detector follows the fundamental of whatever you sing; use the Low–High MIDI selects above the chart to frame the exact register you are training that day, from chest to head voice and mix.

Low—High
Pitch gate9%
Listening — sing any note

Sing with the microphone,
or upload an audio file.

How to Use Pitch Visualization

The graph shows estimated fundamental frequency over time. The horizontal axis uses mm:ss time labels (for example 01:05). The vertical axis is logarithmic so equal musical distances (semitones) look evenly spaced whether you sing low or high. The plot resizes with the page, uses device pixel ratio for a sharp bitmap on retina-class displays, and gives note labels extra margin on narrow screens so the trace stays readable.

Before you see a trace

  • Turn on the microphone from the control strip and allow access. If the trace is empty, raise Pitch detection sensitivity in the plot settings bar above the chart, or sing closer to the mic.
  • Pitch points only appear when the detector finds a stable tone above the sensitivity threshold; noisy or very soft passages may show gaps.
  • If the graph feels too zoomed in or out vertically, widen or narrow the Low–High MIDI range. Turn off Other pitches if weaker partials clutter the display.

Plot settings bar (above the chart)

  • Low / High — Choose the MIDI note span for the vertical axis (logarithmic pitch still). The same window picks which detected pitch counts as the main tone per slice—the loudest peak inside Low–High, not a louder harmonic outside the window. That main pitch drives the highlighted dot, the Controller readout during record/playback, and playback scrubbing. Sound that only registers outside the window may show as out of display range in the strip. Values are remembered for this browser.
  • Other pitches — When enabled, fainter secondary peaks are drawn alongside the main pitch per time slice; when disabled, only the strongest in-range pitch is shown.
  • Highlight — Emphasizes the nearest pitch row, draws a horizontal reference and label for the current main frequency, and (during playback) places a dot on the curve at the playhead. With Highlight on, a short silence or unstable frame can keep the last valid pitch emphasized for a few seconds before it clears.
  • Sensitivity — 0–100% slider (see below). Synced with the app’s pitch detector and what the graph draws.
  • These choices are saved automatically in local storage and restored when you revisit the page.

Pitch detection sensitivity (0–100%)

The slider is labeled in percent (0% = strictest, 100% = most receptive). Internally it maps linearly to the detector: each step changes the thresholds by a predictable amount—no “flat” middle where small moves seem to do nothing.

  • What it controls — (1) Live analysis: live microphone analysis uses a lower RMS gate and autocorrelation peak ratio as you raise the slider. Uploaded files are analyzed once with fixed maximum-receptivity thresholds; moving the slider afterward filters stored candidates without reanalyzing the file. (2) The graph: each time slice keeps up to three detected peaks in history; when drawing, the main pitch per slice is the strongest peak inside Low–High whose relative strength corr ÷ max_corr meets a display floor that also moves linearly with the slider.
  • 0% (strict) — RMS gate ≈ 0.10, peak must reach ≈ 98% of the frame’s best correlation, display floor ≈ 0.98: only very loud, clear fundamentals tend to appear; Other pitches off hides weaker pitch candidates entirely.
  • 100% (receptive) — RMS gate ≈ 0.00008, peak threshold ≈ 12% of max correlation, display floor ≈ 0.10: soft singing, breathy tone, and weaker pitch candidates are much more likely to show.
  • 50% — All three layers sit near the midpoint (for example display floor ≈ 0.54), so half the slider travel matches half the strictness.
  • Default (~9%) — Out of the box the slider reads about 9%, close to the older fixed 0.01 sensitivity for everyday practice.
  • Recorded takes — Peaks are stored in history without a separate amplitude cut at save time. After you stop recording or upload a file, you can move the slider to hide or reveal weaker dots on an existing session without re-analyzing audio.
  • Other pitches — With the checkbox off, only the in-range main pitch per slice is drawn when it passes the display floor; with it on, secondary peaks in that slice are shown too (still subject to the same floor for choosing the “main” highlight).

Browser fullscreen

  • The Fullscreen control sits in the same row as the options above. It requests fullscreen on the main app content: the top header (title, menu, and shared Controller strip) stays on screen with the pitch graph, while the sidebar and other page chrome are hidden.
  • On this detail view, the graph area is padded (similar to the home dashboard) so the trace is not flush against the window edge.
  • Click Exit Fullscreen (same button) or press Esc to leave fullscreen. The plot bar and transport stay usable the whole time.

Recording vs playback

  • While recording, new points are added along the timeline in real time.
  • When you stop recording, the take stays on screen in a frozen snapshot so you can inspect it. Starting a new recording clears the graph for the next session.
  • After you stop recording or playback, a one-line take summary replaces the live readout: median cents vs the nearest tempered note, the share of detected time within ±20¢, and the roughest 5-second stretch — its start time and the average cents off key in that window (time-weighted |¢|). The mm:ss time is a button — click it to move the playhead and progress bar to that moment (same as scrubbing in the Controller). It uses the same main pitch per slice as the graph (Low–High window and Pitch gate).
  • During playback, the chart can follow the playhead and show the note name and frequency at the current time; with Highlight enabled, the playhead marker tracks along the trace.

Chart interaction

  • Auto-follow (default): the time window scrolls so recent audio stays in view.
  • Click and drag horizontally on the plot to pan along time (the vertical range is controlled by Low / High, not by dragging). Dragging usually pauses auto-follow until you turn it back on.
  • Double-click the chart to clear a fixed time window and re-enable auto-follow; a purple on-canvas hint reads “Drag Mode (Double-click to reset)” while auto-follow is off.
  • If you dragged the view earlier, double-click the plot or start a new recording so the window is not stuck off-screen on mobile.

Good use cases

  • Vocal intonation, interval practice, and seeing vibrato width over time.
  • Instrument tuning and comparing long tones across registers.
  • Pair with the Controller for gain, monitoring, file export, and precise scrubbing on the progress bar.

What to watch for on the live trace

  • Spot flat or sharp moments in a phrase while you sing — adjust vowels and support, then repeat.
  • Observe vibrato width and steadiness on sustained notes.
  • Compare pitch stability before and after a warmup routine.
  • Check intonation at the top and bottom of your comfortable range.

How a point gets onto the graph

Short windows of audio are correlated against delayed copies. The delay with the strongest match is treated as the period, then converted to a MIDI note and a cents offset from 12-tone equal temperament. The vertical axis is logarithmic.

The sensitivity slider moves three numbers together, the same ones listed under Pitch detection sensitivity above. At 0% the RMS gate is about 0.10, a peak must reach about 98% of that frame's best correlation, and the display floor is about 0.98. At 100% the RMS gate is about 0.00008, the peak threshold is about 12%, and the display floor is about 0.10. Each time slice keeps up to three peaks. Other pitches, when checked, shows the extras. When it is off, you see the strongest in-range peak that clears the display floor.

Questions

Why is the pitch graph empty?

Turn the microphone on and allow it. If the line is still missing, raise Pitch detection sensitivity, sing closer, or widen Low–High so the note is inside the window.

What are flat and sharp on this graph?

The line is your fundamental against the nearest tempered note. The take summary reports median cents and the share of detected time inside ±20 cents. That ±20¢ figure is this page's summary band. It is not a claim about what every listener can hear.

Why does the line jump an octave?

The detector is autocorrelation: it finds the delay where the waveform best matches a shifted copy of itself. A strong harmonic can win that match, especially on a breathy or very bright tone. Narrow Low–High to the register you are actually singing. Live detection searches roughly 20–5000 Hz and retains up to three candidates; Low–High selects the main displayed candidate, so a narrow window cannot recover a candidate that was not retained.

Is my voice sent anywhere?

No. Detection and drawing run in the tab from the microphone or from a take loaded in the Controller.

The mic button does nothing after I denied the prompt — now what?

The browser remembers that denial for this site until you change it. Open the site's permission settings from the icon in the address bar (a camera/mic icon or a padlock, depending on the browser), set microphone to allow, and reload the page.