Waveform and Spectrogram Viewer, Explained

After reading this you will be able to read an audio file as both a waveform and a spectrogram, compute peak and RMS levels yourself, and spot clipping, silence, hum and bandwidth limits by eye.

What the two views show

An audio file is a list of numbers. Each number is one sample of the sound pressure at one instant. A CD stores 44,100 samples per second, so a 3 second clip holds 132,300 numbers per channel. Two pictures make sense of that list.

The waveform plots sample value against time. The vertical axis is amplitude, usually normalized to the range -1 to +1. You see loudness envelope, silence, and hard clipping at a glance.

The spectrogram plots frequency against time and uses brightness for energy. It answers a different question: not "how loud" but "which pitches are present, and when". A steady 60 Hz hum shows up as a bright horizontal line near the bottom. A cymbal crash shows up as a vertical smear across the top.

Here is the hook. A voice recording sounds fine on cheap earbuds but muddy on studio monitors. The waveform looks normal. The spectrogram shows a solid bright bar at exactly 50 Hz running the whole file: mains hum from a badly grounded cable. You would never see it in the waveform because 50 Hz is quiet compared to speech, but it stacks up on a subwoofer.

When to use it, and when not

Use the waveform when you care about level and timing: is the take clipped, where does the silence start, are two tracks aligned. Use the spectrogram when you care about content by frequency: is there hum, is the top end rolled off, is this file a fake high-resolution copy that was actually upsampled from an MP3.

Do not use either view to judge perceived loudness precisely. Peak amplitude and RMS are physical measures, not loudness. A −14 LUFS master and a −9 LUFS master can share the same peak. For tempo work, a waveform will not give you beats per minute directly; use the BPM Tapper & Detector instead.

The numbers behind the waveform

Two statistics summarize a block of samples. Peak is the largest absolute value. RMS (root mean square) is the energy-like average that tracks perceived level far better than peak.

Here x_i is one sample value in the range -1 to +1, and N is the number of samples in the block. Squaring removes the sign, the mean averages the energy, and the square root returns to amplitude units. Levels are usually reported in decibels relative to full scale:

A full-scale amplitude of 1.0 is 0 dBFS. Half that, 0.5, is 20 \log_{10}(0.5) \approx -6.02 dBFS. Each halving of amplitude subtracts about 6 dB. RMS is always at or below peak, so its dBFS value is always at or below the peak value.

The numbers behind the spectrogram

The spectrogram comes from a short-time Fourier transform (STFT). You chop the signal into short overlapping frames, taper each frame with a Hann window to reduce edge artifacts, and take the discrete Fourier transform of each frame. The magnitude at each frequency bin becomes one pixel of brightness.

The frame length sets the trade-off. With a sample rate f_s and a frame of N samples, each frequency bin is this wide:

At 44100 Hz with a 2048 sample frame, \Delta f = 44100 / 2048 \approx 21.5 Hz per bin. Longer frames give finer frequency resolution but blur time. Shorter frames sharpen time but smear frequency. That is the uncertainty principle for signals, and no window choice escapes it.

The highest frequency any digital signal can carry is the Nyquist frequency, half the sample rate.

At 44100 Hz that is 22050 Hz. The spectrogram cannot show anything above it, and a hard black ceiling below Nyquist is a strong sign of low-pass filtering or lossy compression.

Worked example with the demo clip

Load the demo. It uses the default settings on a short synthetic tone plus noise sampled at 44100 Hz. Work the level math on a small block so you can check every digit.

Reading and interpreting what you see

Clipping

The waveform flattens against a horizontal line at the top or bottom. Peak reads 0.0 dBFS exactly and stays there for many samples. In the spectrogram, clipping adds harmonics, so you see extra bright horizontal lines above the original tone.

Silence and dropouts

A flat line at zero. RMS over that block falls toward -\infty dBFS. A brief drop to silence mid-signal is a dropout, often a corrupt sample or a bad edit.

Hum

A steady horizontal line in the spectrogram at 50 or 60 Hz, often with fainter lines at multiples (100, 150 Hz or 120, 180 Hz). Hard to see in the waveform.

Bandwidth limit

A sharp black cutoff in the spectrogram below the Nyquist frequency. A file at 44100 Hz that goes dark above 16 kHz was almost certainly a lossy MP3 at some point.

Try the frequency and time trade-off

Common mistakes

The first mistake is trusting peak alone. A track can peak at -0.1 dBFS and still sound crushed if its RMS sits at -5 dBFS, a crest factor of only 4.9 dB. Read both.

The second is reading the spectrogram brightness as linear. It is decibels. A faint trace can still be audible, and a bright region is not proportionally louder.

The third is confusing a low sample rate with low quality by looking only at the ceiling. A genuine 44100 Hz recording of a muffled voice may have little energy above 8 kHz because the source had none, not because of compression. Look for a sharp cutoff, not a gentle roll-off. Compression cuts hard; nature rolls off.

Frequently asked questions

Related tools

Song Practice Studio

Drop any song and get its chords, key and tempo in sync with the music, then loop, slow down and transpose it to practise along.

BPM Tapper & Detector

Tap a tempo or detect BPM from an audio file.

Audio Format Converter

Convert between WAV, MP3, OGG, FLAC, M4A and more.

Audio Trimmer

Cut a section out of an audio file, with fades.

Chord Identifier

Name any chord from notes you pick or type.

Euclidean Rhythm Generator

Generate world-music rhythms from two numbers.