What Is the Cepstrum?

The cepstrum is a signal representation obtained by computing the inverse Fourier transform of the logarithm of a signal's power spectrum. Introduced in 1963 by B. P. Bogert, M. J. Healy, and J. W. Tukey in a paper titled "The Quefrency Analysis of Time Series for Echoes," the cepstrum was devised as a tool for detecting echoes and periodically structured patterns in seismic and acoustic data. The independent variable of the cepstrum is called the quefrency, a term the authors created by rearranging the letters of "frequency"; quefrency is measured in units of time, and a peak at a given quefrency value indicates a periodic structure in the log spectrum with a period equal to that time value.

The derivation of the word "cepstrum" from "spectrum," together with the deliberate coinage of a parallel vocabulary, reflected the authors' recognition that they had opened a new analysis domain with its own conceptual structure. As Oppenheim and Schafer document in their historical account published in IEEE Signal Processing Magazine, the idea had wide and rapid uptake, particularly in speech processing, where the relationship between cepstrum and the source-filter model of speech production proved especially productive.

Mathematical Foundation and the Role of the Logarithm

The cepstrum exploits a key property of the logarithm: because convolution in the time domain corresponds to multiplication in the frequency domain, and because the logarithm converts multiplication to addition, the log spectrum of a signal that is the convolution of two components is the sum of the log spectra of those components. The cepstrum, as the inverse Fourier transform of this log spectrum, therefore represents the two components as additive contributions in the quefrency domain. This property makes the cepstrum a natural tool for separating source and filter in signals modeled as the convolution of an excitation and a system response, the central insight behind homomorphic signal processing as developed by Alan Oppenheim.

The power cepstrum, which uses the real-valued inverse Fourier transform of the logarithm of the power spectral density, is most common in applications where phase information is not needed. The complex cepstrum retains phase and admits a reconstruction path back to the original signal, which is required for applications such as echo removal and speech coding. MATLAB's Signal Processing Toolbox documentation on cepstrum analysis provides a practical guide to computing both the real and complex cepstrum and interpreting the quefrency axis.

Relationship to Fourier Analysis

The cepstrum can be understood as a second application of Fourier analysis, with the logarithm inserted between the two transforms. Both the cepstrum and the Fourier transform expose periodicities in their respective input domains: the Fourier transform detects periodicity in the time domain, while the cepstrum detects periodicity in the log spectrum. When a signal contains an echo, the log spectrum exhibits periodic ripples whose frequency in the quefrency domain equals the echo delay. This connection to Fourier transform theory means that engineers already fluent in spectral analysis can apply their intuition about frequency-domain peaks and periodicities directly to cepstral analysis, with the quefrency axis interpreted analogously to the time axis in a standard spectrum.

Applications

The cepstrum and cepstral analysis have applications across a wide range of signal processing domains, including:

• Pitch detection and fundamental frequency estimation in voiced speech and music
• Speech coding and voice synthesis through source-filter separation
• Speaker identification and voice biometrics
• Echo detection and deconvolution in seismic exploration data
• Mechanical diagnostics using vibration cepstra to detect gear faults and bearing defects
• Audio watermarking and forensic audio analysis