Understanding the Spectrogram¶

The Visualizer shows every recording as a spectrogram. This article explains what you are looking at, why the picture stops where it does, and how to read the small line of facts under the player. If you want the click-by-click guide, see Visualize Recordings as Spectrograms.

The three axes¶

A spectrogram shows three things at once:

Time runs left to right. The Visualizer's x-axis is in seconds from the start of the recording.
Frequency runs bottom to top, in kilohertz (kHz). Low sounds (wind, traffic, a bullfrog) sit near the bottom; high sounds (many insects, most bird song, bat calls) sit higher up.
Loudness is the brightness or colour of each point. In the default grayscale, dark means quiet and light means loud. The colour presets (the palette button) map the same numbers to different colours; they do not change the data.

Behind the picture is a long series of short measurements. The audio is cut into overlapping windows a few milliseconds long, each window is analysed for how much energy it has at each frequency (a Fast Fourier Transform, FFT), and each result becomes one thin vertical column of the image.

Why the picture stops at a certain frequency — the Nyquist ceiling¶

Every recording has a sample rate: how many times per second the recorder measured the microphone (48,000 times per second = 48 kHz). A digital recording can only contain frequencies up to half its sample rate. That limit is called the Nyquist frequency.

Recorder sample rate	Highest frequency in the recording
8 kHz	4 kHz
16 kHz	8 kHz
22.05 kHz	11.025 kHz
32 kHz	16 kHz
44.1 kHz	22.05 kHz
48 kHz	24 kHz
96 kHz	48 kHz
192 kHz	96 kHz
250 kHz	125 kHz
384 kHz	192 kHz

This is not a display limit — the sound above the ceiling was never captured. A 48 kHz recording cannot show a 45 kHz bat call, no matter how the spectrogram is drawn. If you need higher frequencies, the recorder has to be configured for a higher sample rate before deployment: see Choosing a Sample Rate for Your Recorder.

Note

The Visualizer's y-axis ends at the recording's Nyquist frequency. Two recordings from different recorders can therefore have different-height spectrograms for the same sound. Analysis forms (Pattern Matching templates, Audio Event Detection frequency bands) are bounded by the same ceiling.

The facts line above the player¶

At the top of the player panel the Visualizer shows a compact, right-aligned line like:

60.000 seconds @ ≤ 24 kHz (48 kHz sample-rate) · 16-bit

Read left to right:

60.000 seconds — the exact duration, to the millisecond.
≤ 24 kHz — the frequency range the recording can contain: the Nyquist ceiling, half the sample rate. Nothing above this exists in the recording.
(48 kHz sample-rate) — the rate the recorder sampled at, which sets that ceiling.
16-bit — the bit depth: how finely each sample's loudness was measured. 16-bit is the norm for field recorders. (Shown when known; many FLAC uploads do not record it.)

Hover any item for a longer explanation. Only facts the system actually knows are shown; nothing is guessed. The stored file size is deliberately not shown — it describes the compressed copy Arbimon keeps, not the recording.

For recordings uploaded after 18 October 2026 the line continues with quality measurements taken at upload time:

peak −6.2 dBFS — the loudest sample, relative to full scale (0 dBFS is the loudest value the file can hold).
floor −58.8 dBFS — the level of the quietest tenth of the recording, excluding pure digital silence: a practical estimate of the noise floor.
clipped 0.08 % — shown only when some samples hit full scale in runs (flat-topped waveforms, which sound distorted). A recorder gain set too high is the usual cause.
silent 94 % — shown only when a large share of the recording is pure digital silence (not a quiet forest — exactly zero), which points at a recorder or SD-card problem.
DC +0.023 — shown only when the whole waveform is shifted away from zero, a recorder or microphone fault signature.

These are measurements, not judgements. A recording can be perfectly useful with a low peak or a high floor; the numbers are there so you can compare recorders and sites.

Reading shapes¶

A horizontal line is a steady tone at one frequency: an insect's constant hum, a generator, electrical interference.
A vertical streak is a short broadband sound: a click, a snap, rain on the housing.
A diagonal sweep is a sound that rises or falls in pitch: many bird notes and bat echolocation calls.
Stacked horizontal bands with equal spacing are harmonics of one source — a frog, a mammal call, a vehicle engine.
A solid dark bar at the top of a 44.1 or 48 kHz recording is simply the region above the Nyquist ceiling if the axis is set wider than the recording can fill. Nothing is missing; nothing could be there.

Gain, filter and playback speed do not change the recording¶

The gain dial makes playback louder; the frequency filter lets you listen to one band; the speed control slows or speeds up playback (useful for bat and insect sound, which can be brought into hearing range by slowing to 10–25 %). None of these alter the stored audio or the spectrogram's data — they change only what you hear.