DSP sound/audio demos

Hi All. 

The last few weeks I have been asked by a few people (neighbours, 
neighbours' kids) what I do for a living. And I find that it's not 
easy to explain verbally what DSP is about. Terms like "Discrete 
Fourier Transforms", "Singular Value Decompositions", "Rayleigh waves" 
and the likes don't do much for lay people, except perhaps establishing 
or reinforcing my already firm reputation as "the mad scientist" of the 
neighbourhood.

So, inspired by this and some remarks Al Clark made in the DSP book thread, 
I've started thinking what demos could be useful to show people what 
at least parts of DSP is about. I assume I have a laptop PC available 
where matlab is installed and that has a microphone and a sound recording 
program. With this system I can record somebody speaking a sentence,
store the sound to a .wav file that I load into matlab. I then manipulate 
this sound by means of DSP in matlab, and play the manipulated result over 
the PC speaker or headphones.

So, what DSP demos could be included here? Remember, I'm not an audio 
specialist so the most psycho-acoustical stuff would probably best be 
left out. Implementations should not be too difficult either, or at 
least good algorithm recipes should be available. Pitch shifting? 
LPC-errors (somebody told me these things sound like "robot voices" 
from a 1977 sci-fi movie)? Adding echos/reverbrations? Other things?

Rune

Rune Allnor wrote:
|| Hi All.
||
|| The last few weeks I have been asked by a few people (neighbours,
|| neighbours' kids) what I do for a living. And I find that it's not
|| easy to explain verbally what DSP is about. Terms like "Discrete
|| Fourier Transforms", "Singular Value Decompositions", "Rayleigh
|| waves"

Hey Rune,
I have been reading a couple of your questions-
They were asking "for a living" not interest or hobby.
If you show off, be aware that there might be a guy in your neighborhood,
who really does this for a living, then you *are* caught.
BTW what are you really doing for a living?

ciao Ban
(from S.E.D and lurking here)

Hello all,

I hope this is not too off-topic for this list - I am assuming that dsp experts 
are more likely to be conversant with analog design than vice versa!

I am writing some introductory pedagogical material (for musicians interested in 
music technology) about digital filters, and I am doing this by starting with 
the technical terms and phenomena associated with analog filters, as used in the 
professional sound studio. So, I am writing about phase response, and the 
inherent 90deg phase shift of capacitors and inductors, hence the general 
non-linear phase response of filters involving poles. Later on, in introducing 
digital filters, I will of course have to cover FIR filters, and the matter of 
linear phase.

I have this irresistible assertion, that analog filters cannot exactly achieve 
linear phase FIR style, so that the FIR filter has no exact counterpart in the 
analog domain. Is this is a correct assertion?  My web searches, etc, have found 
things like Bessel filters which are "almost" linear phase in the passband (I 
assume these are favoured for speaker crossover filters). In short, I seem to 
find this "almost" crops up everywhere. In effect this is also asking if an IIR 
filter can be linear phase, and here too I hit "almost" everywhere I have looked 
so far.

Turning this question around, and making it more general, how accurate a copy 
can the digital filter ever be, of an analog one? I know about the bilnear 
transform and frequency warping; but what else is there that makes life 
difficult for the designer trying to make a copy of a much-loved and 
idiosyncratic analog filter, that will pass the analog-synthesist-junkie test?


Thansk in advance for all answers,

Richard Dobson

Richard Dobson wrote:
|| Hello all,
||
|| I hope this is not too off-topic for this list - I am assuming that
|| dsp experts are more likely to be conversant with analog design than
|| vice versa!

This is something I really doubt!

||
|| I have this irresistible assertion, that analog filters cannot
|| exactly achieve linear phase FIR style, so that the FIR filter has
|| no exact counterpart in the analog domain. Is this is a correct
|| assertion?  My web searches, etc, have found things like Bessel
|| filters which are "almost" linear phase in the passband (I assume
|| these are favoured for speaker crossover filters). In short, I seem
|| to find this "almost" crops up everywhere. In effect this is also
|| asking if an IIR filter can be linear phase, and here too I hit
|| "almost" everywhere I have looked so far.
||
You are too much focused on digital and do not understand analog at all,
despite the above statement. In the analog world filters are *never* linear
phase. Good filters are minimum phase hopefully.
No, crossovers are usually Linkwitz-Riley 4th order nowadays.
Look at it this way: the loudspeaker in itself is already an array of
filters. It is a highpass with a mass-spring mechanism and this arrangement
has what kind of response?
Guess, minimum phase of course and when you want to compensate or modify its
response then you need again minimum-phase filters.
And that is what you find in all these speaker-management-systems. With IIR
you achieve this with minimum length, just simple BiQuads. But it is as well
possible with FIR filters, but for high frequency ratios they get a bit
long...


|| Turning this question around, and making it more general, how
|| accurate a copy can the digital filter ever be, of an analog one? I
|| know about the bilnear transform and frequency warping; but what
|| else is there that makes life difficult for the designer trying to
|| make a copy of a much-loved and idiosyncratic analog filter, that
|| will pass the analog-synthesist-junkie test?
||
A digital filter cannot exactly simulate an analog filter, because of the
limited frequency response. This is especially true with filters close to
the Nyquist frequency.
We might not hear a difference though, but measurement-wise there is a big
difference.
||
|| Thansk in advance for all answers,
||
|| Richard Dobson

ciao Ban

Richard Dobson wrote:

> Hello all,
> 
> I hope this is not too off-topic for this list - I am assuming that dsp 
> experts are more likely to be conversant with analog design than vice 
> versa!
> 
> I am writing some introductory pedagogical material (for musicians 
> interested in music technology) about digital filters, and I am doing 
> this by starting with the technical terms and phenomena associated with 
> analog filters, as used in the professional sound studio. So, I am 
> writing about phase response, and the inherent 90deg phase shift of 
> capacitors and inductors, hence the general non-linear phase response of 
> filters involving poles. Later on, in introducing digital filters, I 
> will of course have to cover FIR filters, and the matter of linear phase.
> 
> I have this irresistible assertion, that analog filters cannot exactly 
> achieve linear phase FIR style, so that the FIR filter has no exact 
> counterpart in the analog domain. Is this is a correct assertion?  My 
> web searches, etc, have found things like Bessel filters which are 
> "almost" linear phase in the passband (I assume these are favoured for 
> speaker crossover filters). In short, I seem to find this "almost" crops 
> up everywhere. In effect this is also asking if an IIR filter can be 
> linear phase, and here too I hit "almost" everywhere I have looked so far.
> 
> Turning this question around, and making it more general, how accurate a 
> copy can the digital filter ever be, of an analog one? I know about the 
> bilnear transform and frequency warping; but what else is there that 
> makes life difficult for the designer trying to make a copy of a 
> much-loved and idiosyncratic analog filter, that will pass the 
> analog-synthesist-junkie test?
> 
> 
> Thansk in advance for all answers,
> 
> Richard Dobson

One or the techniques you need to learn is how to start a new thread.
  	[Request for consensus: should that be in
  	http://users.erols.com/jyavins/procfaq.htm?]
Linear-phase analog filters can't be built with a finite number of
elements using analog techniques, although approximations are available.

Be careful about what you mean by "the inherent 90deg phase shift of
capacitors and inductors". That is indeed the relation of voltage to
current, but input to output is not so constrained.

In theory, you can approximate any phase and frequency response with
either digital or analog filters by using enough elements, allowing a
loose enough approximation, and waiting long enough for the output.

It seems to me that the trend in crossover networks is toward splitting
the signal to feed separate amplifiers. High power at reasonable
impedances requires higher voltages that convenient, and amplifiers are
cheap. Splitting the band to separate amplifiers mitigates that.
Individually powered speakers in arrays of the same kind are also in the
wings. Self-powered speakers of many quality levels are already
available. One DSP can provide all the computation for this, so I expect
R-L crossovers to join the tube amplifiers in my attic.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

In article tbNtb.5$396.4@news-binary.blueyonder.co.uk, Richard Dobson at
richarddobson@blueyonder.co.uk wrote on 11/16/2003 11:01:

> I hope this is not too off-topic for this list - I am assuming that dsp
> experts are more likely to be conversant with analog design than vice versa!

that may or may not be the case, but the topic is definitely ON-topic for
comp.dsp.

> I am writing some introductory pedagogical material (for musicians interested
> in music technology) about digital filters, and I am doing this by starting
> with the technical terms and phenomena associated with analog filters, as used
> in the professional sound studio. So, I am writing about phase response, and
> the inherent 90 deg phase shift of capacitors and inductors, hence the general
> non-linear phase response of filters involving poles. Later on, in introducing
> digital filters, I will of course have to cover FIR filters, and the matter of
> linear phase.
> 
> I have this irresistible assertion, that analog filters cannot exactly achieve
> linear phase FIR style, so that the FIR filter has no exact counterpart in the
> analog domain. Is this is a correct assertion?

i think it is correct, in the semantic context that you are thinking about.
i.e. if you really want to dot your Is and cross your Ts, you should
probably differentiate between the terms "analog" and "continuous-time".

a causal continuous-time filter that doesn't have some sorta perfect delay
elements in it (one constructed from op-amps or other amplification
elements, resistors, capacitors, possibly inductors), that filter cannot be
FIR.  it's impulse response is an additive combination of one-sided
exponential and sinusoidal functions.

now there are (or used to be) these devices that used to be called "bucket
brigade delays" (i think the legit name was "charge-coupled devices" which
operated on sampled analog signals.  that is, the analog signal was sampled
in time (so it became a discrete-time signal rather than a continuous-time
signal) but it was never digitized so it remained an analog voltage that
lived as discrete samples on capacitors that, using some fancy solid-state
switching technology, got transferred from one capacitor to the next in
line.  those capacitors could be wired to high input impedance (FET) op-amps
and the voltages of these could be weighted and summed to a summing
junction.  even though it's done as analog, it is essentially the same
mathematical operation that is an FIR filter.  that would be an "FIR style"
analog filter and could just as well be linear-phase if the weighting
coefficients where symmetric.

>  My web searches, etc, have
> found things like Bessel filters which are "almost" linear phase in the
> passband (I assume these are favoured for speaker crossover filters). In
> short, I seem to find this "almost" crops up everywhere. In effect this is
> also asking if an IIR filter can be linear phase, and here too I hit "almost"
> everywhere I have looked so far.

Bessel filters *are* mostly linear-phase (or constant time-delay) over the
passband frequencies.  because Bessel filters have such sloppy roll-off
(don't make the sharpest cut-off), what is often done is that virtually
phase-linear analog filters are made with Butterworth filters that have
sharper cutoff and with mostly linear-phase *except* for a big bump in the
phase or group delay right around the cut-off frequency.  since this bump is
not really wildly shaped (this DSP book by Grover and Deller describes the
phase response of the elliptical filter as "drunk fly on cross-country skis
in tornado" so don't use elliptical is phase linearity is important to you),
a well designed all-pass filter can compensate for *most* of that bump in
the delay of the Butterworth LPF.  so "almost" is the appropriate word for
continuous-time analog filters that are meant to be phase-linear.
 
> Turning this question around, and making it more general, how accurate a copy
> can the digital filter ever be, of an analog one? I know about the bilnear
> transform and frequency warping; but what else is there that makes life
> difficult for the designer trying to make a copy of a much-loved and
> idiosyncratic analog filter, that will pass the analog-synthesist-junkie test?

if the sampling-rate is good and high (like 96 kHz or higher), i think the
frequency response (both magnitude and phase) of nearly any analog filter
can be well approximated (in the domain of frequencies of interest) by a
digital filter.  there are other issues to worry about to *really* satisfy
the palette of the analog-synthesist-junkie.  these would be non-linearities
and noise.

figuring out exactly what the analog filter is doing as the input signal
gets cranked up might be difficult to do with precision.  and emulating that
would be difficult if you don't first figure it out.  and the ears of the
analog-synthesist-junkie always seems to know the difference between
approximately emulating this behavior and exactly emulating it.

also maybe the analog-synthesist-junkie likes a little low-level noise in
his filters and would miss it if it were gone.  some of this noise *must* be
put in a digital filter in the form of dither if you want to eliminate the
non-linear behavior of signal quantization that inevitably must be done in a
practical implementation.  but if the word size gets larger and larger, that
dither and quantization noise can be made *much* smaller than the natural
resistance and op-amp noise of an analog filter.  maybe, in this case "more
is less".

how about hum?  should digital filter designers try to slip in a little
synthesized 60 Hz hum into their emulations?  maybe there could be a setting
to change it to 50 Hz for the analog-synthesist-junkie that lives on the
other side of the pond.

> Thansk in advance for all answers,

might not be the ones you wanted.  are you doing an article for Electronic
Musician?

r b-j

"Ban" <XbansuriX@XwebX.de> wrote in message news:<lgMtb.29683$9_.1173558@news1.tin.it>...
> Rune Allnor wrote:
> || Hi All.
> ||
> || The last few weeks I have been asked by a few people (neighbours,
> || neighbours' kids) what I do for a living. And I find that it's not
> || easy to explain verbally what DSP is about. Terms like "Discrete
> || Fourier Transforms", "Singular Value Decompositions", "Rayleigh
> || waves"
> 
> Hey Rune,
> I have been reading a couple of your questions-
> They were asking "for a living" not interest or hobby.
> If you show off, be aware that there might be a guy in your neighborhood,
> who really does this for a living, then you *are* caught.

That's true, at least what audio is concerned. There are some people 
nearby who apparently have quite a reputation among professional musicians 
as "wizards" what sound in concert halls and PA is concerned.

My intention by asking this question is to become able to give a
comprehensable explanation to lay people and kids *without* "showing off". 
When I ask for ideas involving audio, it's for preparing explanations more 
along the lines of "I go out and make a measurement of sound somewhere, 
load the recorded sound into my computer and do all sorts of things with 
it". I think that if I elaborate from there in terms of voice data 
(preferably recorded there and then, so that the data gets a more "personal 
touch" for whatever audience present) it may become a bit more 
comprehensable than if I talk about all sorts of arcane seismic or other 
experiments. 

> BTW what are you really doing for a living?

I'm currently a post.doc on DSP and acoustics at the Norwegian University 
of Science and Technology. I wrote a PhD thesis on DSP on seismic data 
at the same university (dissertation September 2000), and my work was 
funded by an oil company. I was involved for a couple of years with an 
underwater acoustics lab. I'm currently looking for work with the seismics 
industry.

The main subject of my work is DSP on, and analysis of, acoustic (though 
not audio) data. 

Rune

robert bristow-johnson wrote:

   ...

> now there are (or used to be) these devices that used to be called "bucket
> brigade delays" (i think the legit name was "charge-coupled devices" which
> operated on sampled analog signals.  that is, the analog signal was sampled
> in time (so it became a discrete-time signal rather than a continuous-time
> signal) but it was never digitized so it remained an analog voltage that
> lived as discrete samples on capacitors that, using some fancy solid-state
> switching technology, got transferred from one capacitor to the next in
> line.  

   ...

Those things are still widely used. CCD imaging arrays shift the pixel
information out that way. Otherwise, we would need an ADC per pixel for
binary coded outputs. Switched-capacitor filters are another example of
discrete-time analog circuits in common use.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

On Sun, 16 Nov 2003 16:01:27 +0000, Richard Dobson
<richarddobson@blueyonder.co.uk> wrote:

>>I have this irresistible assertion, that analog filters cannot exactly achieve 
>>linear phase FIR style, so that the FIR filter has no exact counterpart in the 
>>analog domain. Is this is a correct assertion?  

No.  They are called "transversal filters", and they predate
digital FIR filters.  They use analog delay lines.

>>Turning this question around, and making it more general, how accurate a copy 
>>can the digital filter ever be, of an analog one? I know about the bilnear 
>>transform and frequency warping; but what else is there that makes life 
>>difficult for the designer trying to make a copy of a much-loved and 
>>idiosyncratic analog filter, that will pass the analog-synthesist-junkie test?

I know of a frequency-domain least-squares fitting technique that
models phase as well as amplitude.  I have used it to create
digital approximations for analog filters to within a fraction of
a dB in amplitude and on the order of a degree in phase, IIRC.

Greg Berchin

"Jerry Avins" <jya@ieee.org> wrote in message
news:bp8sn0$cul$2@bob.news.rcn.net...
> robert bristow-johnson wrote:
> Those things are still widely used. CCD imaging arrays shift the pixel
> information out that way. [...]

How does that work?  The only way I can think of needs inductances.