Scribd
Upload
62 views3 pages

Cepstral Analysis: Appendix 3

frequency response

Uploaded by

alialataby
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
62 views3 pages

Cepstral Analysis: Appendix 3

frequency response

Uploaded by

alialataby
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Fundamentuls of Fluid Mechanics and Transport Phenomena

Jean-Laurent Peube
Copyright 02009, ISTE Ltd.

Appendix 3

Cepstral Analysis

A3.1. Introduction

The cepstrum is an integral transform that can be calculated from a spectrum and
contributes to its analysis. For example, in a vibration problem, it makes it possible
to separate the impulse response from the excitation forces. The analysis of vibration
signals from rotating machinery is important. The different operating conditions of a
machine, as well as defects, can be observed in the spectral domain ([STR 96]).

It is also used in speech processing, the vocal signal coming from the
convolution of the excitation (source) and the impulse response of the vocal passage.

Another application is the extraction of incident sound from a signal containing


both the incident sound and its reflection (echo suppression). We will quickly cover
this application, which enables the demonstration of the use of logarithms and thus
of the cepstrum.

A3.2. Definitions

Let x(t) be a time signal and Fx(Q) its Fourier transform; by definition the
complex cepstrum Cx(t) is the inverse Fourier transform, denoted F-1, of the
logarithm of function Fx (Q ) ([BOU 98], [NOR 03 ], [STR 96])

C x (W ) F 1 (ln^Fx (Q )`)

if Fx (Q ) F x (Q ) e jM (Q ) , then: ln^Fx (Q )` ln^Fx (Q ) `  j M (Q )


484 Fundamentals of Fluid Mechanics and Transport Phenomena

The power cepstrum may be defined in several ways, for example, as the square
of the modulus of the complex cepstrum of the signal:

2
C px F 1 (ln^Fx (Q )`) .

or by taking the inverse Fourier transform of logarithm of the modulus square:

C px
©
^
F 1 §¨ ln Fx (Q )
2
`·¸¹
A3.3. Example of echo suppression

Let x(t) be a sound signal comprising the superposition of an original sound s(t)
and an unwanted echo sr(t):

x(t ) s (t )  s r (t )

The reflected sound is attenuated and dephased compared with the original signal

s r (t ) a s (t  t 0 )

The Fourier transform of x(t) can be written:

Fx (Q )
Fs (Q ) 1  a e 2S j Q t0

and the square of the modulus of this transform is:

F x (Q )
2 2

Fs (Q ) 1  a 2  2a cos 2S Q t 0
If we take the logarithm of this quantity, the echo phenomenon is seen in
frequency space by the addition of a periodic term, of period 1/t0:

^
ln Fx (Q )
2
` ln^F (Q ) ` ln 1  a
s
2 2
 2a cos 2S Q t 0
We now take the inverse Fourier transform of this expression ([ALL 04], ([MAD
98], [NOR 03]): the first two terms will be, respectively, the power cepstrums x(t)
and x(t); the third term will be the inverse Fourier transform of the periodic function
shown above, which will be comprised of Dirac functions of different amplitude,
separated by t0
Appendices 485

C px (W ) ^
C ps (W )  F 1 ln 1  a 2  2a cos 2S Q t 0 `
We thus find ourselves in a pseudo-temporal space, whose variable is known as
“quefrency”, in which we perform a “liftering” of the Dirac signals (the terms
“cepstrum”, “quefrency” and “liftering” are respectively anagrams of the words
spectrum, frequency and filtering). The echo signal is thus eliminated by the
suppression of the Dirac signals in this space. The initial signal is thus reconstructed
without its echo by an inverse process ([NOR 03]); however, we note that phase
information has been lost with this procedure of cepstrum power.

A3.4. General case

We consider a source signal x(t) going through a passive linear system whose
impulse response is h(t). The output signal y(t) is the convolution product of x by h:

y (t ) h(t ) * x(t )

The convolution product becomes a simple product after Fourier transform:

F y Q H Q Fx Q

The cepstrum transforms the product to a sum:

C y (W ) C h (W )  C x (W )

In order to perform the deconvolution, in other words the separation of the


source from that of the medium, some assumptions must be made regarding the class
of functions which comprise either the source or the medium:
– in the case of speech, we assume that the vocal excitation is comprised of a
periodic impulse t0 (this property is of course only applied to the stable parts of
sounds) and that the cepstral contribution of the vocal passage is found in the low
quefrencies; we can thus obtain a smoothed spectrum which only contains
information on the vocal passage. The inverse Fourier transform of the spectrum
gives the estimated impulse response of the passage and is used for voice synthesis
([JUR 08], [MAD 98]).
– in mechanics, free oscillatory periods of a medium are generally small
compared to the periods of the excitations: the cepstrum of the impulse response of
the structure is in the left part, while that of the excitation is in the right part of the
cepstrum graph of structure response to excitation ([DES 00], [NOR 03], [ALL
04]).

You might also like