Shepard Tone
EN | FR
In the Timbre Lingo post about Auditory Scene Analysis, we introduced some of the ways that sound components are grouped together by the auditory system. This has many musical applications, and also lays the foundation for some interesting auditory illusions. One of the most famous is the Shepard Tone, named for cognitive scientist Roger Shepard, which creates an illusion of perpetual ascent or descent.
Shepard Tones may take the form of continuous glissandi, as in the example above (these are also called Shepard-Risset Glissandi, after the composer Jean-Claude Risset who pioneered them). Other kinds of Shepard Tones may be presented in scales or melodies of discrete notes, as in this example from the classic video game Super Mario 64.
Shepard Tones are musical “barber shop poles” that exploit some of the cues that our auditory systems use to group sound components. Several frequencies moving in parallel tend to group together (as per the frequency comodulation principle of Auditory Scene Analysis). In a Shepard Tone, octave-related partials move together in an ascending or descending pattern, as is readily visible in the following spectrogram:
As partials reaching the top end of the spectrum disappear, they are slyly replaced by new partials entering at the bottom (or vice versa). The effect is enhanced by dynamic tapering-off of partials at the extreme high and low ends.
Just as regular tones can be combined into different kinds of chords, several Shepard Tones can be superimposed to create interesting harmonic results, such as the augmented and minor triads that may be heard in the first two examples above. Shepard Tone-like musical constructions appear in the works of many 20th- and 21st-century composers, including Risset, Karlheinz Stockhausen, Witold Lutosławski, Georg-Friedrich Haas, Francis Dhomont, and many others. They also figure prominently in timbre perception and cognition research, such as this study on ambiguity and context sensitivity of brightness perception by Kai Seidenburg.
On this page: