Subject: Re: comp.speech Frequently Asked Questions - part 2/3
From: jot@hplb.hpl.hp.com (Jonathan Thompson)
Date: Fri, 20 Dec 1996 15:59:49 GMT
James Salsman (jsalsman@bovik.org) wrote:
: Jonathan Thompson wrote:
: >
: > ...You mean to say that a scaling factor has been added to
: > the definition of the cepstrum for the sake of military secrets.
: No, I mean to say that the cepstrum was redefined for the
: sake of military secrets. What makes you say that the
: difference between DFT(log|DFT(.)|) and IDFT(log|DFT(.)|)
: is a scaling factor?
Yes, If you look in any DSP book you will see the only difference in the
DFT and IDFT algorithms is that the IDFT is scaled by 1/N, where N is the
length of the DFT. So what effect does this have on the cepstrum?
1) The 'Oppenheim-Schafer' coefficients are 1/N smaller than
`Bogert' cepstrum.
2) If you tried to decode `Bogert' cepstra with the `Oppenheim-Schafer'
algorithm, you would get a speech signal with a dynamic range of
10000dB!! So you'd soon realise something was wrong there.
: > I think you'll probably find that the reason for all subsequent
: > references to to the cepstrum using an IDFT instead of a DFT is
: > because the original paper was wrong.
: How can the original use of a novel term be wrong? I've been
: studying this topic for over a year, and I have no doubt that the
: Schafer-Oppenheim cepstrum is very nearly the identity; it's only
: a slight convolution. If you would read Appendix 2 of the
: Bogert, Healy, and Tukey paper, it will be quite clear that their
: definition is rock-solid.
I think the quote in Herve Broulard and Nelson Morgan's book is quite apt -
`nothing is invented and perfected at the same time' - need I say more.
: > ... far more plausible than some conspiracy theory.
: What's implausible about a government at war (c. 1970) and with
: much earlier levels of encryption technology wanting to protect
: cockpit and other high-noise-environment voice radio encryption
: schemes from automated attacks which require a way to determine
: when the plaintext speech has been recovered?
The cepstrum in either guise is not a military secret, they both have been
published!! Having worked on military equipment design (a number of years ago
before joining H.P.) I can tell you that everything that is to be published
has to be approved, so the cepstrum according to your argument would never have
seen the light of day. Also virtually all algorithms, used in military equipment
are fairly standard and can be found in papers and books, it's the way they
use them, and the exact parameters of the system that is secret. If you want
to know how the British military CODEC of that era worked, then see John Holmes
paper on the JSRU filter-bank (IEE Proceedings 1980). There you
will find that it is not the coder that is secret, but the way they
encrypt the filter bank coefficients. The same will apply to the US
cepstrum system (if it exists).
: Could it be that H-P, one of the founders of which was
: Secretary of Defense in that era, might have been involved?
: See 37 USCFR 5.2, http://www.kuesterlaw.com/lawrule/rules9.htm#52
: Sincerely,
: :James Salsman
:
As my signature says, I speak for myself not for H.P. I am not a spokes-
person for the Industrial-Military machine that some people think rules
this world (Oops the X-Files is creeping in again).
To find out why Oppenheim and Schafer define the cepstrum in the way they did,
have you ever considered asking them? E.g. write a letter to Messieurs
Oppenheim and Schafer c/o IEEE Signal Processing Society. I think you'll get
an answer that has some mathematical reasoning behind it, rather than some
blarney about them having to change it at the behest of the DoD.
Jonathan Thompson
--
The views represented here are mine own and not necessarily those of my
employer, Hewlett Packard.
Subject: comlex systems summer school
From: Ismael Santos
Date: Fri, 20 Dec 1996 21:06:53 +0000
Time Evolution in Complex Systems
April 1 to 14, 1997
Summer School, Oeiras, Portugal
This Summer School is part of a program on time evolution in complex
systems which is promoted by the Institute for Scientific
and Technological Advanced Studies - Portugal (ISTAS - Portugal).
Recent progress in the field of dynamic processes in materials is
reported from different theoretical and experimental
perspectives. Theoretical descriptions relate to inhomogeneous systems,
phase transitions, phase ordering dynamics, pattern
formation and time-dependent density functionals. The main experimental
tools are X- ray/neutron diffraction, spectroscopic
and laser techniques. Applications are in the field of advanced material
sciences.
Topics
Statistical mechanics of surfaces and interfaces
Metal surfaces
Defect structures
Strain related microstructures
Computer simulations of mesoscopic structures
Structural phase transitions in metals
Phase transitions and optical spectroscopy
Phase ordering dynamics
Time dependent density functionals
Complex pattern dynamics
Invited Lecturers
A. J. Bray (University of Manchester, UK)
Ulrich Bismayer (Universitaet Hamburg, Germany)
Carlos Fiolhais (Universidade de Coimbra, Portugal)
Alain Gibaud (Université du Maine, France)
Eberhardt Gross (University of Wuerzburg, Germany)
Armen Khachaturyan (Rutgers University, USA) (to be confirmed)
Adam Kiejna (University of Wroclaw, Poland)
Ekhard Salje (University of Cambridge, UK)
Maxi San Miguel (Universidad de las Islas Baleares, Spain)
Franz. Schwabl (Technische Universitaet Muenchen, Germany)
Margarida Telo da Gama (Universidade de Lisboa, Portugal)
Igor. Tsatskis (University of Cambridge, UK)
Wim van Saarloos (Leiden University, The Netherlands)
Organizing Commitee
Ekhard Salje (Cambridge, UK)
Carlos Fiolhais (Coimbra, Portugal)
Correspondence Address
Summer School April 1997
ISTAS - Portugal
Quinta da Nora, Apartado 3028
3000 Coimbra
Portugal
Tel: +351-39-700937
Fax: +351-39-700912
E-Mail: istas@ipn.uc.pt
Format
Two or three lectures will be held each day, together with informal
discussions and poster presentations by the participants.
Important Deadlines
Application for Grants: January 15, 1997
Summer School registration: February 15, 1997
Abstract submission: March 15, 1997
Grants
A number of grants will be available for participants from EU countries.
These grants will be preferentially given to young
participants (graduate students, post-docs, etc.) Applications including
a short C.V. (with a list of publications) should be sent
to the address given above. The deadline for grant application is
January 15, 1997.
The applicants will be notified before February 15, 1997 by E-mail or
fax whether their applications were successful.
Registration
Please complete the enclosed registration form and return it to the
Summer School secretariat before February 15, 1997.
Fees
Please follow the payment instructions given on the form.
The registration fees for the Summer School are as follows:
Regular, before February 15, 1997 US dollars 300
Regular, after February 15, 1997 US dollars 400
Submission of Abstracts
Contributions on the Summer-School topics will be accepted for
presentation in poster form. Authors should submit an
abstract to the program committee. The deadline is March 15, 1997.
They will be notified before March 31, 1997 by E-mail or fax whether
their abstracts have been accepted.
Social Program
A number of social events for participants and their companions will be
offered throughout the Summer-School period (e.g.,
dinner and visits to historical sites in and near Lisbon).
Hotel Accommodation
Tentative advance reservations will be made by the organizing committee.
Participants should complete the corresponding
information in the registration form.
Oeiras and how to reach it
Oeiras is located by the river Tejo about 10 km west from Lisbon. There
are railway connections from Cais do Sodré station
in Lisbon to Oeiras. The organizing committee will arrange bus transport
from and to the Lisbon Airport.
Maps of the region and city of Oeiras
Summer School site
The meeting will be held at the Instituto de Tecnologia Química e
Biológica in Oeiras.
Summer School sponsors
This Summer School is financed by the the European Commission and ISTAS
- Portugal and cosponsored by:
Câmara Municipal de Oeiras
European Science Foundation
Fundação Calouste Gulbenkian
Fundação Luso Americana para o Desenvolvimento
Instituto de Tecnologia Química e Biológica
Ministério da Ciência e Tecnologia
ISTAS - Portugal, istas@ipn.uc.pt
Oct 22, 1996
--
_______________________________________________________________________
Name: Ismael Santos
Organization: I.S.T.A.S.- Portugal (Intitute for Scientific
and Tecnological Advanced
Studies - Portugal)
I.S.T.A.S. Home Page: Http://www.ipn.uc.pt/~istas
I.S.T.A.S. E-mail Address: istas@ipn.uc.pt
Personal Home Page: http://members.tripod.com/~IsmaelSantos/index.html
Personal E-mail Address: Ismaail @mail.telepac.pt
_______________________________________________________________________
Subject: Re: comp.speech Frequently Asked Questions - part 2/3
From: James Salsman
Date: 20 Dec 1996 21:01:48 GMT
Jonathan Thompson wrote:
> James Salsman (jsalsman@bovik.org) wrote:
> : Jonathan Thompson wrote:
> : >
> : > ...You mean to say that a scaling factor has been added to
> : > the definition of the cepstrum for the sake of military secrets.
> :
> : No, I mean to say that the cepstrum was redefined for the
> : sake of military secrets. What makes you say that the
> : difference between DFT(log|DFT(.)|) and IDFT(log|DFT(.)|)
> : is a scaling factor?
>
> Yes, If you look in any DSP book you will see the only difference in the
> DFT and IDFT algorithms is that the IDFT is scaled by 1/N, where N is the
> length of the DFT.
If that were the case, why would FFT-IFFT code libraries have
similar-length definitions for each function, instead of a
simple loop to rescale the outputs after a call to the other
more lengthy function?
I've been through FFT code and chapter 33 of _Intro._to_Algorithms_
by Cormen, Leiserson, and Rivest. I believe there is a grain
of truth here, but the scalling in question occurs deep inside
the inner loop of some implementation of the {I,}DFT, and does
not apply simply to the outputs.
Would you please cite (or better yet excerpt) a source showing
this difference in scaling between the DFT and IDFT?
Sincerely,
:James Salsman
Subject: Re: comp.speech Frequently Asked Questions - part 2/3
From: James Salsman
Date: Fri, 20 Dec 1996 17:29:27 -0800
Jonathan Thompson wrote:
>
> James Salsman (jsalsman@bovik.org) wrote:
> : ... What makes you say that the
> : difference between DFT(log|DFT(.)|) and IDFT(log|DFT(.)|)
> : is a scaling factor?
>
> ... If you look in any DSP book you will see the only difference in the
> DFT and IDFT algorithms is that the IDFT is scaled by 1/N, where N is the
> length of the DFT.
This is partly true of the magnitude of the outputs,
but not for the phase of the outputs, which are
negated between the DFT and IDFT.
> So what effect does this have on the cepstrum?
A lot, because the cepstrum is a complex vector,
because you need to specify the phase of the sines
summing to the envelope in addition to their respective
magnitudes in order to describe that envelope.
More could be said about the quefrency of the harmonic
of the voice excitation. But I've said it already, so
use DejaNews or something.
Sincerely,
:James Salsman
Subject: * TRACS: Training and Research on Advanced Computing Systems
From: desplat@epcc.ed.ac.uk (J-C Desplat)
Date: Thu, 19 Dec 1996 17:23:41 GMT
**********************************************************************
**********************************************************************
** **
** **
** TRACS: Training and Research on Advanced Computing Systems **
** **
** **
**********************************************************************
**********************************************************************
Edinburgh Parallel Computing Centre is coordinating an EC-funded
project to bring European researchers for short visits to associated
departments in Edinburgh to collaborate on projects involving High
Performance Computing (HPC). The TRACS programme is funded by the
Comission of the European Communities under the Training and Mobility of
Researchers Programme.
TRACS provides:
* opportunities to visit and work in Edinburgh
* access to a wide range of HPC systems, including
+ Cray T3D (512 alphas)
+ Cray J90 (10 processors)
+ Sun/SGI Workstation cluster
* training, support and consultancy on parallel computing
* accommodation, travel and subsistence expenses
TRACS is open to both academic and industrial researchers who are
nationals of, and working in, an EC member state or an Associated State
(Norway, Iceland and Liechenstein). Applications will be reviewed by a
selection panel and will be approved on the basis of scientific merit.
For more information on EPCC and the TRACS programme, see
http://www.epcc.ed.ac.uk. Application forms and further information are
available both in electronic (dvi/ps) and paper formats and can be
obtained from:
TRACS Administrative Secretary
EPCC
University of Edinburgh
JCMB, The King's Buildings
Edinburgh EH9 3JZ
United Kingdom
Tel: +44 131 650 5986
Fax: +44 131 650 6555
Email: TRACSadmin@ed.ac.uk
The closing date for applications to be considered at the February selection
meeting is 29 January 1997.
--
Dr Jean-Christophe DESPLAT - Applications Scientist (TRACS)
Edinburgh Parallel Computing Centre - the University of Edinburgh
JCMB King's Buildings - Mayfield Rd - Edinburgh EH9 3JZ - SCOTLAND
Phone: +44 (0)131 650 6716 - Fax: +44 (0)131 650 6555
E-mail: desplat@epcc.ed.ac.uk - http://www.epcc.ed.ac.uk/~desplat
Subject: Re: comp.speech Frequently Asked Questions - part 2/3
From: robert@audioheads.com (robert bristow-johnson)
Date: Fri, 20 Dec 1996 23:26:31 GMT
let's make a silly issue a little bit sillier:
In article , Jonathan Thompson (jot@hplb.hpl.hp.com) writes:
>James Salsman (jsalsman@bovik.org) wrote:
>: Jonathan Thompson wrote:
>: >
>: > ...You mean to say that a scaling factor has been added to
>: > the definition of the cepstrum for the sake of military secrets.
>
>: No, I mean to say that the cepstrum was redefined for the
>: sake of military secrets. What makes you say that the
>: difference between DFT(log|DFT(.)|) and IDFT(log|DFT(.)|)
>: is a scaling factor?
>
>Yes, If you look in any DSP book you will see the only difference in the
>DFT and IDFT algorithms is that the IDFT is scaled by 1/N, where N is the
>length of the DFT.
that is not the _only_ difference. even if you redefined the
DFT to have a scaling factor of 1/sqrt(N) (something i might
like to see since it preserves scaling symmetrically) so that
both the DFT and IDFT have the same scaling, there is still a
difference. one has a "j" in it and the other has a "-j".
that difference is, of course, qualitative. the roles of j and
-j can be reversed and all the rules still apply as well since
they both square to be -1. in fact, on the planet Zork (one of
those places Spaceman Spiff visits) they might have picked the
imaginary number we call "-j" to be their (positive) imaginary
unit. both definitions are equally valid because both j and -j
square to be -1 which is the only salient property of the
imaginary unit (what we call "j"). then what we call a "forward
Fourier Transform", they would call an "Inverse Fourier
Transform" and vise versa. big deal. if they are consistant
with their own definition (of j and -j), everything works out
to be the same (imaginary reality is still the same).
however, if on the same planet, you change the definition as
someone has done above, something (besides scaling) really does
get changed. the minus sign has to go somewhere to make two
comparable definitions.
> So what effect does this have on the cepstrum?
> 1) The 'Oppenheim-Schafer' coefficients are 1/N smaller than
> `Bogert' cepstrum.
>
> 2) If you tried to decode `Bogert' cepstra with the `Oppenheim-Schafer'
> algorithm, you would get a speech signal with a dynamic range of
> 10000dB!! So you'd soon realise something was wrong there.
and 3) the order of your cepstrum data is reversed because
of above.
>My god the X-files has a lot to answer for.
along with Oliver Stone.
r b-j
wave mechanics, inc.
robert@audioheads.com or robert@wavemechanics.com
and, no, i do not believe in the "Single Bullet Theory".
