Spectrum analyzer

The fundamental is the leftmost tall ridge while you speak: the pitch of your voice. Most adult male voices sit roughly between 85 and 180 Hz, most adult female voices roughly between 165 and 255 Hz. Hold a steady "ahh" and you will see it park in that range.

Harmonics are the evenly spaced ridges marching up to the right of the fundamental, at exact multiples of it. They are where a voice's character lives: a 120 Hz fundamental puts harmonics at 240, 360, 480 Hz and on up. Two voices with the same pitch sound different because their harmonics have different strengths.

Sibilance shows up only on "s" and "sh" sounds, as short bursts of energy roughly between 4 and 10 kHz, centered near 6 kHz for many voices. Say "sixty six silver spoons" and watch the right half of the plot light up on every s.

The noise floor is the carpet that remains when you stay silent. A flat, low carpet is broadband hiss from the mic's own electronics or a fan. A sharp spike at 50 or 60 Hz, often with partners at 100 or 120 Hz, is mains hum from a ground loop or a cheap power supply, and the analyzer will name the exact frequency if you hover it.

Once you can see the regions, the obvious next question is what to do about each one. That is exactly what an equalizer is for, and what an EQ is walks through it on the same 20 Hz to 20 kHz axis you are looking at here. For a band by band map of where rumble, boxiness, presence, and air live, keep the EQ frequency chart open next to the analyzer.

A neat experiment: open the tone generator in another tab, play a quiet 1 kHz sine through your speakers, and watch it appear as a single narrow spike here. Switch the wave to square and the odd harmonics at 3, 5, and 7 kHz pop into view. That is the whole idea of a spectrum in one picture.

The frequency axis is logarithmic. Each octave (a doubling of frequency) gets equal width, which matches how pitch is heard: 100 to 200 Hz is the same musical step as 5 to 10 kHz. On a linear axis the entire voice range would be squashed into the first tenth of the plot.

dBFS is digital level, not loudness in the room. 0 dBFS is the loudest a digital signal can be, and every value below it is negative. The plot cannot tell you sound pressure in dB SPL because that depends on your mic's sensitivity and gain, which the browser does not know. What the plot is good at is comparisons: this peak is 20 dB above that one, the hum sits 30 dB above the surrounding floor.

FFT size is a resolution versus speed trade. Frequency resolution is the sample rate divided by the FFT size, so at 48 kHz a 2048-point FFT resolves about 23 Hz per bin and reacts in about 43 ms, while 8192 resolves about 6 Hz but each frame covers about 171 ms of audio and the display visibly lags. Use 2048 to watch speech move, 8192 to pin down a hum's exact frequency. The stepped look at the low end of the plot at 2048 is that resolution limit made visible, not a bug.

Honest caveats. The dB values come from the browser's AnalyserNode, a windowed FFT referenced to full scale, so a single tone can read a few dB low from windowing. The sample rate shown is the audio context's rate, which is what the browser resampled your mic to, not necessarily the hardware rate. And the trace only shows what your mic picks up: a quiet plot on a laptop mic does not mean your room is silent, it can mean the mic is insensitive.

If the silent-room carpet on your plot sits high, or a 50 or 60 Hz spike never goes away, that noise is riding under everything you record. It is fixable in real time, and how to reduce background noise on your mic covers the gate plus suppression recipe step by step. For a single-number read of your floor instead of the full picture, the free mic test measures it in five seconds.