Audio Spectrum Analyzer

How to use the spectrum analyzer

1. Press Start microphone and allow access when your browser asks. The spectrum starts moving as soon as sound comes in.
2. To analyse a recording instead, drop an audio file on the drop zone (MP3, WAV, OGG, FLAC…) and press Play file. You hear the file while it is analysed.
3. Pick a View: Bars, Line or Spectrogram (waterfall), which scrolls from right to left and shows how the spectrum changes over time.
4. Choose the Frequency axis: Logarithmic (like the ear hears, one octave takes the same width everywhere) or Linear (equal hertz per pixel, useful for harmonics and high frequencies).
5. Adjust the FFT size and the Smoothing to trade detail for stability (see below).
6. Press Freeze to hold the current picture and read it calmly, then Resume. Stop releases the microphone and closes the audio engine.

The readouts under the graph give the Peak frequency (the loudest component above 20 Hz), the Nearest note with its deviation in cents (A4 = 440 Hz), the Peak level and the Resolution of the analysis.

What the analyzer shows

A spectrum analyzer splits a sound into its frequencies with a Fast Fourier Transform (FFT). Each block of samples becomes a list of “bins”, and bin k sits at k × sample rate ÷ FFT size hertz. The height of each bar is the energy in that band, in decibels relative to digital full scale. The browser applies a Blackman window before the FFT, so a pure full-scale sine wave reads around −14 dB rather than 0 dB: the values are best used to compare frequencies with each other, not as an absolute sound level meter.

The peak frequency is refined between two bins with parabolic interpolation, so it is more precise than the bin spacing alone. It shows the strongest component, which is usually, but not always, the note you hear: on a low bass note or a voice, the second harmonic can be louder than the fundamental.

FFT size, resolution and smoothing

The FFT size sets the frequency resolution: at 48 kHz, 2048 points give about 23 Hz per bin, 8192 give 5.9 Hz and 32768 give 1.5 Hz. A larger FFT separates close frequencies and low notes better but needs a longer slice of audio (32768 samples is about 0.7 s), so it reacts more slowly to fast changes. For speech and drums, 1024 to 4096 works well; for tuning, bass notes or hum hunting, use 16384 or more.

Smoothing averages each frame with the previous ones. At 0 you see every flicker; around 0.8 the display is calm and easier to read. The spectrogram uses the same data, with colours from black (quiet) through purple and red to yellow and white (loud).

Practical uses

• Find a mains hum (50 or 60 Hz and its harmonics) or a whistling ground loop.
• Check which frequencies an instrument, a voice or a room resonance emphasises before you touch an equaliser.
• Test a speaker or headphones with a sweep from a tone generator and look for dips and peaks.
• See the cut-off of an MP3 file: many low-bitrate encodes stop around 16 kHz.