Seismic Infrasound Viewer

1. Read the reference chart first. It is the most reliable part of this tool and needs no microphone. Each row shows the approximate frequency band where a geophysical or atmospheric event radiates infrasound energy, with cited published values.
2. (Optional) Start the live view. Press Start live view and grant microphone access. The waterfall and spectrum cover 0–20 Hz. The tool reads the real AudioContext sample rate and shows the resulting frequency resolution.
3. Understand what you are seeing. On almost all consumer hardware, the sub-20 Hz region will show only noise floor — the low-frequency rolloff and OS high-pass remove genuine infrasound before it ever reaches the FFT. The dominant-bin readout and “band-overlap” label are for exploration, not detection.
4. Use Freeze to inspect. Pause the scrolling waterfall to study a moment without the display moving. Press Resume to continue.
5. Stop when finished. Pressing Stop releases the microphone immediately. Closing the tab also stops all capture — nothing is recorded or uploaded.

The reference bands are the takeaway

Infrasound — sound below the ~20 Hz threshold of human hearing — is generated by some of the largest processes on Earth. Because the atmosphere absorbs very little energy at these frequencies, infrasound from eruptions, large meteors and explosions can travel thousands of kilometres. The reference chart shows roughly where each source's energy concentrates: the persistent ocean microbarom peak near 0.2 Hz (a ~5-second period), broadband volcanic and earthquake energy reaching well below 1 Hz, meteor signals processed in roughly the 0.2–3 Hz band, the infrasonic part of thunder peaking around 1–3 Hz, and avalanche/landslide energy near 1–8 Hz.

Why the live view almost certainly shows nothing real

The browser AudioContext samples at typically 44,100 or 48,000 Hz, so the Nyquist limit (the highest capturable frequency) is around 22,050 or 24,000 Hz — that part is fine for infrasound. The problem is the low end: built-in MEMS microphones in laptops and phones roll off steeply below roughly 50–100 Hz, and the operating system or browser typically applies a high-pass filter and automatic gain control that remove sub-20 Hz content entirely. So any energy you see in the 0–20 Hz waterfall is almost always noise floor or a digital artifact, not real infrasound. There is also no barometer in a browser, so the “atmospheric pressure correlation” that real infrasound work relies on is conceptual only here.

The classifier is a band label, not a detection

When the live view runs, the tool finds the strongest bin below 20 Hz and reports which reference band it lands in. This is a band-overlap label, not a claim that an earthquake, eruption or meteor is occurring — several sources share the same frequency range, and on consumer hardware the “dominant” bin is usually just where your device's noise floor happens to peak. Genuine detection requires calibrated microbarometers deployed in arrays, the approach used by research observatories and the CTBTO International Monitoring System (0.02–4 Hz, 60 stations).

The live view streams audio from your microphone and runs a Fast Fourier Transform (FFT) on each short window. To get usable resolution in the infrasound band, the tool requests the largest FFT the Web Audio API allows (32,768 samples). At a 48 kHz sample rate that gives about 1.46 Hz per bin (48000 ÷ 32768) — the actual figure is read at runtime and shown under Frequency resolution. Only the bins below 20 Hz are mapped onto the 0–20 Hz display; bin 0 (the DC offset) is ignored so a constant bias in the signal is not mistaken for a 0 Hz tone.

The waterfall spectrogram stacks successive spectra as rows, newest on top, scrolling downward; brightness encodes relative level. The instantaneous spectrum below it draws the current frame with dashed cyan markers at the centre of each reference band so you can see where, say, the microbarom peak or the avalanche band would fall. The number of waterfall columns is bounded to the canvas width so the display can never grow without limit.

The microphone source is connected to an AnalyserNode only — never to the audio output — so there is no feedback path and nothing is played back. We request the raw signal with automatic gain control, noise suppression and echo cancellation off, because those features reshape the spectrum; even so, a browser or OS may apply low-frequency filtering it does not expose, which is the main reason genuine infrasound rarely survives to the FFT. All buffers are fixed-length, the analysis loop pauses when the tab is hidden, and the microphone and audio context are released the instant you press Stop or close the tab. Audio is analysed live and discarded — never recorded or transmitted.