Voice Pitch Analyzer

Why pitch is better understood as a pattern

A live Hz number changes too quickly to summarize a complete utterance. The analyzer therefore collects reliable voiced frames and describes their distribution. Median F0 gives a robust centre, average F0 provides a second view, and P10–P90 identifies the typical lower and upper region without allowing the most extreme ten percent of frames on either side to define the whole sample.

Pitch movement converts that robust span into semitones, a logarithmic unit that corresponds more closely to musical distance than raw Hertz. Stability describes how concentrated the pitch pattern was. Voiced duration and confidence help reveal whether the browser captured enough clear periodic sound. Read all of these together: one number alone can hide an expressive contour, a weak signal, or a brief octave error.

The curve and statistics describe the sample you made, not a fixed property. Conversation, reading, sustained vowels, acting, and singing produce different patterns. Decide what task you are studying before interpreting movement or comparing two sessions.

A practical guide to each pitch metric

Median and average can be close in a balanced sample, but they diverge when a few unusually high or low frames pull the mean. P10 and P90 deliberately ignore the outermost observations, making the displayed range more resistant to coughs, consonant transitions, detector errors, and brief edge pitches. The range is not the absolute lowest and highest sound you can produce; use the dedicated vocal range test for that separate task.

Movement and stability are related but not interchangeable. A passage can cover a moderate span in a smooth, controlled contour, while another sample can jump unpredictably inside a similar span. This lightweight browser tool summarizes distribution rather than performing linguistic prosody analysis, so it does not identify stress patterns, sentence types, emotions, or dialect features.

• Median F0: robust centre of reliable voiced pitch frames.
• Average F0: arithmetic centre, more sensitive to brief excursions.
• P10–P90: typical pitch band rather than absolute extremes.
• Pitch movement: typical band expressed as semitone distance.
• Stability and confidence: sample-level context, not personal ratings.

Use pitch analysis in a repeatable practice workflow

Choose one question per session. To study conversational movement, read the same paragraph naturally. To study a sustained sound, hold the same comfortable vowel without forcing it. To compare two speaking approaches, record each approach two or three times and preserve the setup. Mixing different texts and tasks makes it impossible to know whether a changed metric came from practice or simply from different vowels and emphasis.

Record in the same room and device position, then note the complete result rather than only median Hz. Several sessions showing a similar change are more informative than repeatedly testing until one preferred number appears. This site does not save history, so keep a private written log if longitudinal tracking is useful to you.

Use measurements to direct listening rather than replace it. Play back your own recordings with a local recorder if you need perceptual comparison, and consider qualified coaching for individual goals. A browser pitch curve cannot evaluate clarity, comfort, expression, technique, or whether a change is sustainable.

Recording conditions and common pitch-detection errors

The YIN method searches each frame for a repeating period. Background speech, music, strong reverberation, clipping, weak volume, breathy phonation, creak, and rapid transitions can make that period ambiguous. Harmonics may occasionally cause an octave-high estimate, while subharmonics can produce an octave-low estimate. Filtering reduces these errors but cannot eliminate them from every consumer microphone recording.

Use a quiet room, moderate level, and a microphone distance of roughly one hand. Avoid whispering and do not shout. If the graph contains implausible jumps or confidence is low, make a fresh clear sample rather than interpreting the anomaly. Device processing differs, so comparisons between a laptop, phone, wired headset, and Bluetooth microphone are not controlled experiments.

Scope, privacy, and safe interpretation

Fundamental frequency is only one part of a voice. This analyzer does not directly measure formants, articulation, vocal-tract anatomy, placement, breathiness, nasality, clinical resonance, emotion, identity, or health. It cannot diagnose a condition, verify a speaker, or determine gender identity, biological sex, age, nationality, accent, or personality.

Audio frames and derived measurements remain in browser memory during the session. They are not uploaded, stored by the site, added to an account, or used to train a model. Record only yourself or someone who has clearly consented, and never use these measurements for high-impact or eligibility decisions.

Comfort takes priority over data. Do not force an unusually high or low pitch to move a graph. Stop if you feel pain or strain. Persistent hoarseness, difficulty speaking, or an unexplained lasting change should be discussed with an appropriately qualified healthcare or speech-language professional.

Illustrative result, troubleshooting, and fit

For example, a short fixed passage might show a median F0 of 182 Hz, a P10–P90 range of 154–214 Hz, 5.2 semitones of movement, and high confidence. The numbers are deliberately illustrative rather than a benchmark. Read them as a set: the median describes the centre, the range describes the usual spread, movement describes pitch distance, and confidence describes the quality of the captured evidence. A spontaneous story or a noisy phone recording would reasonably produce a different pattern.

When a curve contains sudden octave jumps, start by repeating the sample with clear voiced speech and less background noise. Check for a distant microphone, clipping, Bluetooth processing, breathiness, creaky voice, or a second speaker. If the same unusual result repeats under controlled conditions, record the context instead of forcing a preferred number. This analyzer suits learners, speakers, and people comparing their own recordings; it is not suitable for diagnosis, authentication, hiring, access decisions, or classifying another person.