DSP books

Hi,

I have just bought these books:

"Understanding Digital Signal Processing", by Richard G. Lyons
"Introduction to Signal Processing", by Orfanidis
"A Wavelet Tour of Signal Processing", by Stephane Mallat

I must say that I really like Richard's book.
It is very accessible and informative.
The others are a little hard to understand mathematically.

I am thinking about buying the books below also, so my question is:
Does anyone know these books and can they be recommended?

"Wavelets and Filter Banks", by Gilbert Strang, Truong Nguyen
"Practical Time-frequency Analysis: Gabor and Wavelet Transforms with an
Implementation in S", by Rene Carmona, et al
"Digital Audio Restoration", by S.J. Godsill, P.J.W. Rayner

My interest is audio and what I would like to get from the books is
practical ideas and methods that I can program.
"Wavelets and Filter Banks" has allready been recommended, so I almost
certainly will buy that one.

Best regards,
Michael

One of the most useful I've found for Wavelets and a general 
background in functional analysis as applied to DSP is:

   Wavelets and Subband Coding
   Martin Vetterli and Jelena Kovacevic
   Prentice Hall
   ISBN 0-13-097080-8

I must say that I have not found wavelets particularly 
useful for audio.  After a wavelet tranform, you can't 
really do much of anything in terms of coeficient 
manipulation and retain the alias cancellation that provides 
perfect reconstruction in multi-rate systems and there is no 
equivalence to convolution.  Only the Fourier sinusoidal 
basis provides that.

They have mostly been useful for data reduction of video 
information and for other kinds of encoding.  I'm not sure 
if anyone has tried to apply the same methods to audio but 
my suspicion is that the artifacts, while ok for video and 
fingerprints, would be horrible for audio.

I started studying them because intuition told me that 
therein might be a good heuristic solution to the audio 
problem of time/frequency scaling.  If there is, it has 
escaped me.  :-)


Bob

-- 

"Things should be described as simply as possible, but no 
simpler."

                                              A. Einstein

"Bob Cain" <arcane@arcanemethods.com> wrote in message
news:cm4n72015kp@enews2.newsguy.com...
>
>
> One of the most useful I've found for Wavelets and a general
> background in functional analysis as applied to DSP is:
>
>    Wavelets and Subband Coding
>    Martin Vetterli and Jelena Kovacevic
>    Prentice Hall
>    ISBN 0-13-097080-8
>
> I must say that I have not found wavelets particularly
> useful for audio.  After a wavelet tranform, you can't
> really do much of anything in terms of coeficient
> manipulation and retain the alias cancellation that provides
> perfect reconstruction in multi-rate systems and there is no
> equivalence to convolution.  Only the Fourier sinusoidal
> basis provides that.
>
> They have mostly been useful for data reduction of video
> information and for other kinds of encoding.  I'm not sure
> if anyone has tried to apply the same methods to audio but
> my suspicion is that the artifacts, while ok for video and
> fingerprints, would be horrible for audio.
>
> I started studying them because intuition told me that
> therein might be a good heuristic solution to the audio
> problem of time/frequency scaling.  If there is, it has
> escaped me.  :-)
>
>
> Bob
>
> -- 
>
> "Things should be described as simply as possible, but no
> simpler."
>
>                                               A. Einstein

Hi Bob

Thanks, I'll have a look at that book.

So far I have done two things with audio using FFT:
1) Noise reduction and other effects in the frequency domain.
2) Calculation of note frequencies and durations.

Doing 2) I also ran into the time/frequency problem with the FFT.
Therefore I want to learn about wavelets in order to have better
frequency resolution at lower frequencies.
In this situation I am not doing the inverse transform, so I will not
have problems with artifacts.

Michael

On 2004-11-01 08:05:02 +0100, Bob Cain <arcane@arcanemethods.com> said:

> I must say that I have not found wavelets particularly useful for 
> audio.  After a wavelet tranform, you can't really do much of anything 
> in terms of coeficient manipulation and retain the alias cancellation 
> that provides perfect reconstruction in multi-rate systems and there is 
> no equivalence to convolution.  Only the Fourier sinusoidal basis 
> provides that.

No, I don't agree. It's probably not as intuitive as the DFT because 
you don't really have the simple & easy concept of a "frequency", but 
you can use the DWT for audio purposes as well if you know how to the 
DWT relates time and frequency to scale space domain.

Of course, you'll easily run into aliasing problems if you use a 
critically sampled wavelet transform, but that's more because people 
usually approach the DWT as simplistically as the DFT (by "zeroing out" 
parts of the transform for example, to get an idea what happens). This 
isn't a wise thing to do for the DWT and you'll easily be disappointed 
because it's likely that all you'll get is lots of noise.

If you know how to get your operators right you can get as sonically 
pleasing results as with the DFT. To be honest, I find the results of 
(over)applying DFT-based processing (such as subtractive denoising and 
data reduction/perceptual coding) more annoying than the occasional 
"graininess" of DWT based processing.

But that's probably because I'm almost "allergic" to the MP3 artifacts 
these days :-)

> They have mostly been useful for data reduction of video information 
> and for other kinds of encoding.  I'm not sure if anyone has tried to 
> apply the same methods to audio but my suspicion is that the artifacts, 
> while ok for video and fingerprints, would be horrible for audio.

We do it all the time at the company I'm working for. Not with a 
"simple" straightforward DWT, but the theory is similar.
-- 
Stephan M. Bernsee
http://www.dspdimension.com

On 2004-11-01 18:47:36 +0100, Stephan M. Bernsee <spam@dspdimension.com> said:

> We do it all the time at the company I'm working for.

Sorry - that would be http://www.prosoniq.com in case you're interested.

Btw., some audio examples of non-DFT based DSP processing are here: 
http://www.prosoniq.com/html/morphaudioex.html and here: 
http://www.prosoniq.com/html/mpexaudioex.html

And some other related links of interest: http://neuron.prosoniq.com 
and http://mpex.prosoniq.com
-- 
Stephan M. Bernsee
http://www.dspdimension.com

On Sun, 31 Oct 2004 22:02:19 +0100, "DSP curious" <dsp@iname.com>
wrote:

>Hi,
>
>I have just bought these books:
>
>"Understanding Digital Signal Processing", by Richard G. Lyons
>"Introduction to Signal Processing", by Orfanidis
>"A Wavelet Tour of Signal Processing", by Stephane Mallat
>
>I must say that I really like Richard's book.
>It is very accessible and informative.
>The others are a little hard to understand mathematically.
>
>I am thinking about buying the books below also, so my question is:
>Does anyone know these books and can they be recommended?
>
>"Wavelets and Filter Banks", by Gilbert Strang, Truong Nguyen
>"Practical Time-frequency Analysis: Gabor and Wavelet Transforms with an
>Implementation in S", by Rene Carmona, et al
>"Digital Audio Restoration", by S.J. Godsill, P.J.W. Rayner
>
>My interest is audio and what I would like to get from the books is
>practical ideas and methods that I can program.
>"Wavelets and Filter Banks" has allready been recommended, so I almost
>certainly will buy that one.
>
>Best regards,
>Michael

Hi Michael,

   Thanks for your "positive" comments 
on comp.dsp recently about my DSP book.
I appreciate it.

Michael, if you tell me which edition, and which 
"printing Number" of my book that you have, I'll 
send you an errata.  The first edition is the book 
with the red & white cover.  The 2nd edition has 
a red & orange cover.

Which edition do you have?  (The Printing Number 
information is different in the two editions.)

I tried to send you a private E-mail, but it 
"bounced back" to me.

Best Regards,
[-Rick-]

  

DSP curious wrote:

> In this situation I am not doing the inverse transform, so I will not
> have problems with artifacts.

Analysis is where they shine so they may well be worth your 
study.


Bob
-- 

"Things should be described as simply as possible, but no 
simpler."

                                              A. Einstein

Stephan M. Bernsee wrote:


>> They have mostly been useful for data reduction of video information 
>> and for other kinds of encoding.  I'm not sure if anyone has tried to 
>> apply the same methods to audio but my suspicion is that the 
>> artifacts, while ok for video and fingerprints, would be horrible for 
>> audio.
>  
> We do it all the time at the company I'm working for. Not with a 
> "simple" straightforward DWT, but the theory is similar.

Glad to hear that at least my intuition was on the mark, if 
not my ability to see any further than that.  Have you 
published anything or does it all remain proprietary?

Which of your products employ this?


Thanks,

Bob
-- 

"Things should be described as simply as possible, but no 
simpler."

                                              A. Einstein

On 2004-11-02 05:22:38 +0100, Bob Cain <arcane@arcanemethods.com> said:

> Glad to hear that at least my intuition was on the mark, if not my 
> ability to see any further than that.  Have you published anything or 
> does it all remain proprietary?
> 
> Which of your products employ this?
> 
> 
> Thanks,
> 
> Bob

Not sure if my last posting made it to the group, here's the link 
information again:

On 2004-11-01 18:58:09 +0100, Stephan M. Bernsee <spam@dspdimension.com> said:

> Sorry - that would be http://www.prosoniq.com in case you're interested.
> 
> Btw., some audio examples of non-DFT based DSP processing are here: 
> http://www.prosoniq.com/html/morphaudioex.html and here: 
> http://www.prosoniq.com/html/mpexaudioex.html
> 
> And some other related links of interest: http://neuron.prosoniq.com 
> and http://mpex.prosoniq.com
> 

-- 
Stephan M. Bernsee
http://www.dspdimension.com

Bob Cain wrote:
...
> They have mostly been useful for data reduction of video 
> information and for other kinds of encoding.  I'm not sure 
> if anyone has tried to apply the same methods to audio but 
> my suspicion is that the artifacts, while ok for video and 
> fingerprints, would be horrible for audio.

Markus Erne's phd thesis at ETH Zurich might be of interest to you:
"Signal adaptive audio coding using wavelets and rate optimization"

Regards,
Andor