Local Audio Analysis and Recognition
What pitch is my microphone hearing?
Play one sustained, clear pitch near your microphone after you start the tuner. This tool estimates its frequency and nearest equal-tempered note entirely in your browser, with a signal-estimate band describing how suitable the input actually was.
Answer
Microphone audio is analyzed locally in your browser only. It is never uploaded, cached, logged, or included in analytics, exports, or a share URL, and processing stops immediately when you press Stop.
Want to learn more? Tuning, Cents, Temperament, and Reference Pitch
Understand what the band does and does not measure
"Signal estimate: High" means the recent readings agreed closely with each other and the underlying period estimate was clean, not that your instrument is perfectly calibrated or that a whole performance is in tune. "Low" or no result means the current input does not yet meet that bar, whether because of silence, background noise, more than one note sounding at once, or clipping. See the Methodology page for this tool's full Recognition Contract summary.
Supported scope and limitations
This tool is built for one clear, sustained, predominantly monophonic pitched sound between roughly 70 and 1,200 Hz, with a stable window of at least a quarter second and adequate input level. It is not designed to separate simultaneous notes, is not tested for speech, and does not identify what instrument is playing. A stable High-band estimate does not prove intonation quality across an entire performance and does not diagnose an instrument's calibration.
Method
After you start the tuner, the browser requests microphone permission, downmixes the input to mono, and analyzes 2,048-sample frames at your device's actual audio sample rate with a YIN-style time-domain estimator that searches periods corresponding to 70-1,200 Hz. Frames with low signal level, clipping, or no clean period estimate are rejected rather than guessed at. The primary result is the median of the latest five accepted estimates, converted to the nearest 12-tone equal-tempered note using your chosen A4 reference. The signal-estimate band is High, Moderate, or Low based on how many recent estimates were retained and how tightly they agree, never a percentage.
What each state actually tells you
Silence produces no result rather than a stale-looking guess: the tool would rather say nothing than assert a pitch it cannot support. A noisy or polyphonic signal, two or more notes sounding together, is a supported-but-degraded case: the estimator may still lock onto one dominant period, but it is not designed to separate simultaneous pitches, and the band drops accordingly so you are not misled into thinking two notes were correctly resolved into one. Clipping (the loudest samples flattening at the top of the input range) actively degrades the underlying period estimate, so a clipping warning always takes priority over a confident-looking band; the fix is to reduce input gain or move the source away from the microphone, not to trust the reading anyway.
A frequency well below the supported 70 Hz floor, a very low bass note or a room hum, may still produce a numeric estimate from the underlying period search, but that estimate sits outside the contract's validated range and should not be read as a reliable in-range reading. The tool's scope statement exists specifically so a plausible-looking number never substitutes for a validated claim.