Also covariances

This material is revised and expanded from the appendix of draft versions of a recent conference submission, for my own reference. I used (deterministic) correlograms a lot in that, and it was startlingly hard to find a decent summary of their properties anywhere. Nothing new here, but… see the matrial about doing this in a probabilistic way via Wiener-Khintchine representation and covariance kernels which lead to a natural probabilistic spectral analysis.

Ning Ma, autocorrelogram of a processed audio signal

\[\renewcommand{\var}{\operatorname{Var}} \renewcommand{\dd}{\mathrm{d}} \renewcommand{\pd}{\partial} \renewcommand{\bb}[1]{\mathbb{#1}} \renewcommand{\vv}[1]{\boldsymbol{#1}} \renewcommand{\mm}[1]{\mathrm{#1}} \renewcommand{\mmm}[1]{\mathrm{#1}} \renewcommand{\cc}[1]{\mathcal{#1}} \renewcommand{\ff}[1]{\mathfrak{#1}} \renewcommand{\oo}[1]{\operatorname{#1}} \renewcommand{\gvn}{\mid} \renewcommand{\II}[1]{\mathbb{I}\{#1\}} \renewcommand{\inner}[2]{\langle #1,#2\rangle} \renewcommand{\Inner}[2]{\left\langle #1,#2\right\rangle} \renewcommand{\finner}[3]{\langle #1,#2;#3\rangle} \renewcommand{\FInner}[3]{\left\langle #1,#2;#3\right\rangle} \renewcommand{\dinner}[2]{[ #1,#2]} \renewcommand{\DInner}[2]{\left[ #1,#2\right]} \renewcommand{\norm}[1]{\| #1\|} \renewcommand{\Norm}[1]{\left\| #1\right\|} \renewcommand{\fnorm}[2]{\| #1;#2\|} \renewcommand{\FNorm}[2]{\left\| #1;#2\right\|} \renewcommand{\argmax}{\mathop{\mathrm{argmax}}} \renewcommand{\argmin}{\mathop{\mathrm{argmin}}} \renewcommand{\omp}{\mathop{\mathrm{OMP}}}\]

Consider an \(L_2\) signal \(f: \bb{R}\to\bb{R}.\) We frequently overload notation and refer to as signal with free argument \(t\), so that \(f(rt-\xi),\) for example, refers to the signal \(t\mapsto f(rt-\xi).\) We write the inner product between signals \(t\mapsto f(t)\) and \(t\mapsto f'(t)\) as \(\inner{f(t)}{f'(t)}\). Where it is not clear that the free argument is, e.g. \(t\), we annotate it \(\finner{f(t)}{f'(t)}{t}\).

The correlogram \(\cc{A}:L_2(\bb{R}) \to L_2(\bb{R})\) maps signals to signals. Specifically, \(\mathcal{A}\{f\}\) is a signal \(\bb{R}\to\bb{R}\) such that

\[\mathcal{A}\{f\}:=\xi \mapsto \finner{ f(t) }{ f(t-\xi) }{t}\] This is the covariance between \(f(t)\) and \(f(t-\xi).\) (Note that we here discuss the covariance between given deterministic signals, not between two stochastic sources; covariance of stochastic processes is a broader, let alone inferring the covariance of stochastic processes.) Note also that this is what I would call an autocovariance not an auto-correlation, since it’s not normalized, but I’ll stick with the latter for now since for reasons of convention.

We derive the properties of this transform.

Multiplication by a constant. Consider a constant \(c\in \bb{R}.\)

\[\begin{aligned}\mathcal{A}\{cf\}(\xi)&= \inner{ cf(t) }{ cf(t-\xi) }\\ &= c^2\finner{ f(t) }{ f(t-\xi) }{t}\\ &= c^2\mathcal{A}\{f\}(\xi).\\ \end{aligned}\]

Time scaling:

\[\begin{aligned}\mathcal{A}\{f(r t)\}(\xi) &=\finner{ f(r t) }{ f(r t-\xi) }{t}\\ &= \int f(r t)f(r t-\xi)\dd t\\ &= \frac{1}{r }\int f(t)f(t-\frac{\xi}{r})\dd t\\ &= \frac{1}{r} \mathcal{A}\{f\}\left(\frac{\xi}{r}\right)\\ \end{aligned}\]


\[\begin{aligned}\mathcal{A}\{f+f'\}(\xi) &=\finner{ f(t)+f'(t) }{ f(t-\xi)+f'(t-\xi) }{t}\\ &=\finner{ f(t) }{ f(t-\xi)\rangle+\langle f(t),f'(t-\xi) }{t} +\finner{ f'(t) }{ f(t-\xi)\rangle+\langle f'(t),f'(t-\xi) }{t}\\ &= \mathcal{A}\{f\}(\xi)+ \finner{ f'(t) }{ f(t-\xi)}{t} +\finner{f(t)}{f'(t-\xi) }{t} +\mathcal{A}\{f'\}(\xi).\\ &= \mathcal{A}\{f\}(\xi)+ \finner{ f'(t) }{ f(t-\xi)}{t} +\finner{f(t+\xi)}{f'(t) }{t} +\mathcal{A}\{f'\}(\xi).\\ &= \mathcal{A}\{f\}(\xi)+ \finner{ f'(t) }{ f(t-\xi)}{t} +\finner{f'(t) }{f(t+\xi)}{t} +\mathcal{A}\{f'\}(\xi).\\ \end{aligned}\]

We can say little about the term \(\finner{ f'(t) }{ f(t-\xi)}+\finner{f'(t) }{f(t+\xi)}{t}\) without more information about the signals in question. However, we can solve a randomized version. Suppose \(S_i, \, i \in\bb{N}\) are i.i.d. Rademacher variables, i.e. that they assume a value in \(\{+1,-1\}\) with equal probability. Then, we can introduce the following property:

Randomised addition:

\[\begin{aligned} \bb{E}[ \mathcal{A}\{S_1f + S_2f'\}(\xi) &=\bb{E}[ \mathcal{A}\{S_1f\}(\xi) + \finner{ S_2 f'(t) }{ S_1 f(t-\xi)}{t} +\finner{S_2f'(t) }{S_1 f(t+\xi)}{t} +\mathcal{A}\{S_2f'\}(\xi)]\\ &=\bb{E}[ \mathcal{A}\{S_1f\}(\xi)] + \bb{E}\finner{ S_2 f'(t) }{ S_1 f(t-\xi)}{t} + \bb{E}\finner{S_2f'(t) }{S_1 f(t+\xi)}{t} +\bb{E}[ \mathcal{A}\{S_2f'\}(\xi)]\\ &=\mathcal{A}\{f\}(\xi)+ \bb{E}[ S_1S_2]\finner{ f'(t) }{ f(t-\xi) }{t} + \bb{E}[ S_1S_2]\finner{ f'(t) }{ f(t+\xi) }{t}+\mathcal{A}\{f'\}(\xi)\\ &=\mathcal{A}\{f\}(\xi)+ \mathcal{A}\{f'\}(\xi)\\ \end{aligned}\]


Abrahamsen, Petter. 1997. β€œA Review of Gaussian Random Fields and Correlation Functions.”
Bochner, Salomon. 1959. Lectures on Fourier Integrals. Princeton University Press.
Brown, Judith C., and Miller S. Puckette. 1989. β€œCalculation of a β€˜β€˜Narrowed’’ Autocorrelation Function.” The Journal of the Acoustical Society of America 85 (4): 1595–1601.
Cariani, P. A., and B. Delgutte. 1996. β€œNeural correlates of the pitch of complex tones. I. Pitch and pitch salience.” Journal of neurophysiology 76 (3): 1698–1716.
CheveignΓ©, Alain de, and Hideki Kawahara. 2002. β€œYIN, a Fundamental Frequency Estimator for Speech and Music.” The Journal of the Acoustical Society of America 111 (4): 1917–30.
Kaso, Artan. 2018. β€œComputation of the Normalized Cross-Correlation by Fast Fourier Transform.” PLOS ONE 13 (9): e0203434.
Khintchine, A. 1934. β€œKorrelationstheorie der stationΓ€ren stochastischen Prozesse.” Mathematische Annalen 109 (1): 604–15.
Langner, Gerald. 1992. β€œPeriodicity Coding in the Auditory System.” Hearing Research 60 (2): 115–42.
Lewis, J.P. 1995. β€œFast Template Matching.” In. Quebec City, Canada: Canadian Image Processing and Pattern Recognition Society,.
Licklider, J. C. R. 1951. β€œA Duplex Theory of Pitch Perception.” Experientia 7 (4): 128–34.
Loynes, R. M. 1968. β€œOn the Concept of the Spectrum for Non-Stationary Processes.” Journal of the Royal Statistical Society. Series B (Methodological) 30 (1): 1–30.
Ma, Ning, Phil Green, Jon Barker, and AndrΓ© Coy. 2007. β€œExploiting Correlogram Structure for Robust Speech Recognition with Multiple Speech Sources.” Speech Communication 49 (12): 874–91.
Morales-Cordovilla, J. A., A. M. Peinado, V. Sanchez, and J. A. Gonzalez. 2011. β€œFeature Extraction Based on Pitch-Synchronous Averaging for Robust Speech Recognition.” IEEE Transactions on Audio, Speech, and Language Processing 19 (3): 640–51.
Rabiner, L. 1977. β€œOn the Use of Autocorrelation Analysis for Pitch Detection.” IEEE Transactions on Acoustics, Speech, and Signal Processing 25 (1): 24–33.
Slaney, M., and R. F. Lyon. 1990. β€œA Perceptual Pitch Detector.” In Proceedings of ICASSP, 357–360 vol.1.
Sondhi, M. 1968. β€œNew Methods of Pitch Extraction.” IEEE Transactions on Audio and Electroacoustics 16 (2): 262–66.
Tan, L. N., and A. Alwan. 2011. β€œNoise-Robust F0 Estimation Using SNR-Weighted Summary Correlograms from Multi-Band Comb Filters.” In 2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 4464–67.
Wiener, Norbert. 1930. β€œGeneralized Harmonic Analysis.” Acta Mathematica 55: 117–258.

No comments yet. Why not leave one?

GitHub-flavored Markdown & a sane subset of HTML is supported.