How It Works

The Web Audio API is a high-level browser interface for processing and synthesizing audio. Unlike simple audio playback, it provides a node-based graph architecture where audio sources, processors, and destinations are connected in a chain. Snap Decibel Meter creates a minimal graph: your microphone feeds into an AnalyserNode, which performs a Fast Fourier Transform (FFT) on the incoming signal in real time.

The FFT decomposes the raw time-domain waveform into individual frequency components, producing a spectrum that shows how much energy exists at each frequency. From this spectrum, we extract multiple features — total energy (for dB estimation), the balance between low and high frequencies (for sound classification), and the overall shape of the spectrum (to distinguish tonal sounds from broadband noise). All of this happens within a single animation frame, typically every 16 milliseconds, without ever storing the underlying audio.

Sound is a pressure wave — a rapid alternation between compression and rarefaction of air molecules. A microphone converts this pressure variation into an electrical signal, which the phone's analog-to-digital converter samples thousands of times per second. The resulting digital signal oscillates around zero, with the size of the oscillations corresponding to the loudness of the sound.

Simply averaging these samples would give a value near zero because positive and negative swings cancel out. Instead, we use the Root Mean Square (RMS) method: square each sample (making all values positive), calculate the mean of the squared values, then take the square root. The result is a single positive number that represents the effective amplitude of the signal — a well-established technique used in both audio engineering and electrical power measurement.

Converting RMS amplitude to decibels involves a logarithmic formula: dB = 20 × log₁₀(RMS / reference). The logarithmic scale compresses the enormous range of human hearing — from the threshold of perception to the threshold of pain, sound pressure varies by a factor of over one million — into a manageable range of roughly 0 to 140 dB. Snap Decibel Meter works in dB relative to digital full scale (dBFS). Without calibration there is no absolute acoustic reference, so the number on screen is a signal level and is labelled as such rather than as a sound pressure level. The optional one-point calibration adds a fixed offset you derive from a reference meter, which turns the display into an estimated A-weighted level — an estimate on consumer hardware, not a measurement traceable to a physical standard.

Sound classification in Snap Decibel Meter uses a rule-based spectral analysis approach rather than machine learning. The frequency spectrum from the FFT is divided into three bands: low (below 500 Hz), mid (500 Hz–2 kHz), and high (above 2 kHz). The relative energy distribution across these bands creates a rough "fingerprint" characteristic of different sound types, and the label shown is a best-guess estimate rather than a verified identification.

Traffic noise, for example, tends to concentrate energy heavily in the low and low-mid bands, with a gradual roll-off above 1 kHz. Human speech shows distinct energy peaks in the 300 Hz–3 kHz range with characteristic modulation patterns. Music typically has broader spectral content with stronger tonal peaks. Nature sounds often feature high-frequency energy from birdsong combined with broadband low-frequency content from wind or water.

The classifier also examines spectral flatness — a measure of how evenly energy is distributed across frequencies. White noise has high spectral flatness (energy everywhere), while a single musical note has very low flatness (energy concentrated at one frequency and its harmonics). Combined with the spectral centroid — the "center of gravity" of the spectrum, indicating whether the sound is dominated by low or high frequencies — these features allow a surprisingly effective broad classification without the computational overhead or data requirements of neural networks.

Raw audio analysis produces values that fluctuate rapidly — dozens of times per second. Displaying these raw values directly would create a jittery, unreadable gauge. Snap Decibel Meter applies exponential smoothing to the dB output, giving recent samples more weight while gradually incorporating older ones. This creates the smooth, responsive gauge movement you see, which responds quickly to changes in sound level without nervously jumping on every micro-fluctuation.

The peak hold feature captures the highest instantaneous reading during your session and displays it separately from the current level. This is important because many real-world noise concerns involve intermittent peaks — a truck braking, a dog barking, construction impact noise — that are brief but significant. The running average, meanwhile, tells you about the sustained background level, which is more relevant for assessing long-term exposure risk or general ambient noise conditions.