This brief introduction explains how to read the spectrograms (also called sonograms) that can be found on many pages of this website. Some spectrograms are animated, while others are static images. In both cases, the following animation will be helpful in the reading and contains suggestions on the interpretation of the sound spectrum of a signal. For a gallery of spectrograms, look through the drawings or go directly to the pages on musical instruments.

Use the ◀ and ▶ buttons (or the arrow keys after clicking a button) to step through the slideshow. The text of each callout is repeated below the picture.

◀ 1 / 10 ▶

1. The vertical axis is the time axis. Zero is at the top and time runs downwards. The unit is the millisecond.

2. As time goes by, this reference line moves down, and marks the instant to which the spectrum refers.

3. This number gives the milliseconds elapsed from the start to the position of the reference line.

4. This number gives the frequency at which the cursor is.

5. This is the frequency axis (in Hz).

6. The frequency axis also applies to this box, in which the spectrum of the sound is drawn instant by instant.

7. Note: in this box the vertical axis no longer shows time, but the amplitude of the components of the spectrum. The amplitude is measured on a dB scale whose zero is the maximum value.

8. As the reference line moves in time, the spectrum of the sound changes. The upper box shows the spectra at all times... ...while the lower one shows only the spectrum at the present instant.

9. In this case the spectrum consists of an irregular low-amplitude signal, due to the background noise, and of a series of well-separated peaks, regularly spaced in frequency.

10. The colour scale relates the amplitudes of the spectrum in the lower box to the amplitudes of the sonogram in the upper box.

What a spectrogram shows

• A periodic sound, such as a note played or sung, contains only frequencies that are multiples of its fundamental frequency (see Fourier's theorem). In the spectrum it therefore appears as a series of equally spaced peaks, and in the spectrogram as a series of lines parallel to the time axis: the frequency distance between two neighbouring lines equals the fundamental frequency, that is, it gives the pitch of the note. For example, peaks at 220, 440, 660 Hz... indicate a note of 220 Hz.
• When the pitch changes, as in the sung scale of the example, all the lines move together and their spacing changes.
• A noise has no distinct lines: its energy is spread over a wide band of frequencies and in the spectrogram it appears as a uniformly coloured area (see Sound or noise?).
• The colour, or the height of the peaks in the spectrum, shows how strong each component is: the relative amplitudes of the harmonics largely determine the timbre.

The analysis window

To calculate the spectrum of a sound, a program such as Audacity takes a stretch of the signal (the window) of a certain number of samples and calculates its Fourier transform. The result is an average over that stretch. Changing the size or type of the window changes the way the samples are weighted. So the spectrum also changes: it is not an error, they are different measurements. The spectrogram repeats this calculation over successive windows, and the choice of window determines its appearance.

• Resolution in frequency and in time. With the usual sampling rate (44100 samples per second) a window of 512 samples lasts about 12 ms, one of 2048 samples about 46 ms. The shorter window separates frequencies at least a few hundred Hz apart: two peaks at 200 and 400 Hz appear as a single "hill". The longer window separates them well, but follows rapid changes of the sound less well. One resolution cannot be improved without worsening the other (see frequency-time uncertainty). For the voice, windows of a few tens of milliseconds are reasonable.
• Comparisons. To compare different spectra it is best always to use the same window (size and type).
• Sounds that change. The spectrum of a long stretch, such as a whole word, is an average and says nothing about the order in time: a sound and the same sound played backwards have the same amplitude spectrum. To follow the evolution over time you need the spectrogram.
• Negative values in dB. The zero of the scale corresponds to the maximum the system can represent, so a peak at −15 dB is not "inaudible": it is 15 dB below the maximum. The absolute level depends on how the signal was recorded.
• Other algorithms. Autocorrelation, offered by some programs as an alternative to the spectrum, measures something else: how similar the signal is to itself shifted in time. It is useful for estimating the period (and hence the pitch) of a sound, but it does not say which frequencies it contains.

Check your understanding

Further reading and links

• Related questions and answers:
  • How do you read a spectrogram?