Formant Analyzer

Drag & Drop or Click to Select a Recording of Speech

One voice, close to the microphone: WAV, MP3, M4A, Ogg, Opus, or FLAC

Tip: paste an audio file from clipboard (Ctrl/⌘+V).

Formant tracksHz / time

Vowel chartF2 across, F1 down

Every voiced frame as a point; the ring is the median. Reference vowels are Peterson and Barney's male averages.

Shareable link

Settings are written to the URL as you change them. Nothing differs from the defaults yet.

Oxyd To put the result on a track with other audio, Oxyd is a multitrack editor with the same effects. Open Oxyd

Find the resonances of a voice. Each frame of the recording is fitted with a linear predictor, a short all-pole filter that reproduces the frame’s spectral envelope, and the filter’s poles are read out as frequencies and bandwidths. That is how Praat measures formants, and the defaults here are Praat’s, so a figure read off this page lands where Praat’s does on the same file.

The tracks

Voiced frames are plotted as points, one track per formant, F1 brightest. Gaps are unvoiced frames, where no pitch was found and the fit describes noise rather than a vowel. A steady vowel draws five flat lines; speech draws the curves of the tongue moving between them.

The vowel chart

F2 across and F1 down, both on a log axis and F2 reversed, the way phoneticians draw it so that the chart matches the mouth: front vowels on the left, open vowels at the bottom. The grey symbols are the Peterson and Barney male averages from 1952, which are the usual landmarks; the ring is the file’s median. A recording of one held vowel lands as a cloud near its symbol, and a different speaker shifts the whole cloud.

Settings

Highest formant sets the band the fit looks at and the rate the file is resampled to, twice that frequency. It is the control to get right: a ceiling too low loses F5 into F4, one too high invents a formant between the real ones. Formants sets how many poles are fitted, two per formant. The window is 25 ms at Praat’s default, 15 for faster speech, 40 for a steadier reading on a sustained note.

The table

CSV or JSON, one row per 10 ms hop: time, f0, voicing strength, then frequency and bandwidth for each formant, with a 0 where none was found. Rows pool pairwise past 20,000, and the hop reported at the top of the file is the one that applies after pooling.

Limits

A high-pitched voice has harmonics spaced wider than the formants are, so the predictor sees peaks at harmonics and the fit scatters; Praat has the same trouble on children and sopranos. Two voices at once defeat it entirely. Breathy or whispered speech has formants but no pitch, so those frames are plotted as unvoiced even when the table has numbers for them.

Frequently Asked Questions

What is a formant?

A resonance of the vocal tract. The first two, F1 and F2, are what make a vowel that vowel: a high F1 is an open mouth, a high F2 is the tongue forward. F3 and above carry speaker and consonant information. They are frequencies of the tract, not of the voice, so they stay put when the pitch moves.

How are they measured?

The way Praat's To Formant (burg) does it: the recording is resampled to twice the highest formant wanted, pre-emphasised from 50 Hz, cut into Gaussian-windowed frames, and each frame is fitted with a linear predictor by Burg's method. The predictor's roots are the resonances; their angles are the frequencies and their radii the bandwidths.

Which settings match Praat?

The defaults are Praat's: 5 formants, a 5500 Hz ceiling, a 25 ms window hopping 10 ms. Praat recommends 5000 Hz for a man's voice and 5500 for a woman's; a child's or a soprano's needs higher. The ceiling matters more than any other setting, since the fit only looks below it.

Why do unvoiced frames have formants?

The predictor fits whatever is in the frame, and a fricative or silence has resonances of a sort. Each row carries the frame's pitch and a voicing strength so those rows can be dropped; the plots already leave them out.

Explore Our Tools

Browse all tools

Voice Quality Report (Jitter, Shimmer, HNR)

Measure a sustained vowel the way Praat's voice report does: jitter local, RAP and PPQ5, shimmer local, APQ3, APQ5 and APQ11, the harmonics-to-noise ratio, pitch statistics and the fraction of unvoiced frames, as a table, JSON or CSV.

Pitch Detector

Track the fundamental frequency of a recording over time with the YIN difference function, and read the median pitch as a note name, a MIDI number, and its deviation in cents.

Audio Formant Shifter

Move the resonances of a voice up or down by ±12 semitones and 50 cents while every partial stays at its original frequency, so the pitch does not change.

Spectrogram

Plot how an audio file's spectrum changes over time, with an FFT size from 512 to 8192, a log or linear frequency axis, an adjustable floor, and a PNG export.

Audio Feature Extractor

Measure an audio file frame by frame: RMS, zero-crossing rate, spectral centroid, spread, rolloff, flatness, flux and 13 MFCCs, as CSV or JSON with a time column, for notebooks, classifiers and plots.

Audio Spectrum Analyzer

Plot the average magnitude spectrum of an audio file with peak hold, on a log or linear frequency axis. Set the FFT size from 1024 to 16384 samples and read the level of any band in dBFS.

Reverb Time Analyzer (RT60)

Measure a room's reverberation time from a recorded clap, balloon pop or impulse response: T20, T30 and EDT per octave band from 63 Hz to 8 kHz by backward integration, the ISO 3382 method, with the decay curve, the fit and how far each band sits above the noise floor.

Audio Quality Analyzer

Read an audio file's spectral cutoff, clipped sample runs, DC offset, noise floor, and true peak in one pass, and compare the cutoff against the frequency each encoder usually stops at.

Chord Detector

Read the chords of a song from an audio file: a timeline of major and minor triads with their start times, the changes snapped to the beat, a chart by bars with the key on top, and the chords exported as markers for Oxyd or Audacity. Reads the chromagram, not a MIDI file.

Stereo Field Analyzer

Read an audio file's phase correlation, mid and side levels, left-right balance, and per-band correlation, with a goniometer scatter and the level a mono sum would cost.

Impulse Response Capture

Measure a room's impulse response with a sine sweep: a sweep plays through the speakers, the microphone records the answer, the recording is deconvolved to the room's response and saved as a WAV, with its decay, frequency response and mid-frequency reverberation time read on the spot.

Key Detector

Find the musical key of an audio file from its chromagram, with the Camelot code, the confidence, the relative and parallel alternatives, and the full 24-key score table.