On Tue, Feb 14, 2012 at 8:30 AM, Ken Norton <ken@xxxxxxxxxxx> wrote:
> > Probably my short remarks are not helpful for anyone ... there are a
> number
> > of audiophiles on the list and I thought they would see similarities
> > removing sound distortions ... somewhat similar thought process as is
> done
> > digitally with images.
>
>
> There really are similarities. The entire concept of
> expose-to-the-right, is based on the same concept of maximizing
> signal-to-noise in audio. Crank the input to as close to the clip
> point as possible without clipping. Then during the mixdown you pull
> the volume down to the proper level. Inotherwords, always record
> hotter so you pull the levels down, never up.
>
> Other similarities involve high-pass and low-pass filtering. Curves
> adjustment is just like using audio compression or expansion.
>
> Noise filtering in audio is an interesting beast, though. The general
> concept is to sample the noise pattern and then apply a subtractive
> algorithm to the sound to remove the noise pattern. Some algorithms do
> this better than others, and most require a multi-pass approach to do
> well.
>
> There are some interesting studies that show the direct coorelation
> between colors and notes. By applying audio spectrum to visible
> spectrum, you get clashing colors in the same manner than notes will
> clash. I started looking at that in 1994 when a researcher for some
> university bought some equipment from me to test it.
>
> Back to noise filtering. Pattern noise in audio and image data is very
> easy to address. But the problem with both is that they cannot
> efficiently change with the changing image/sound scape. For example,
> the sound of an idling car can be removed. But at the point where the
> driver puts the car in gear and accelerates, the sound would reappear
> since the signature has changed. This is why most noise removal
> algorithms look at the overall dynamic range of the noise, get some
> basic pattern infromation and then apply a broad brush approach to the
> removal. Another highly successful method is to identify what it is
> that you want to keep. By identifying the "real" sound, you are able
> to subtract out what it is that doesn't look or sound like the real
> sound.
>
> Unfortunately, in the case of audio noise reduction, there has been
> very little significant development in that area in over 15 years. The
> algorithms have been improved a bit, but we're pretty much stagnent in
> that subject. In fact, we've actually seen the reduction in tools
> since a couple companies that had effective NR systems are no longer
> in existance.
>
> AG
>
Applying filters etc to equalize frequency response or correct phase
distortion was very much my frame of mind.
The part I made incomprehensible was that a PSF (point spread function) for
an imaging system is very analgous to the impulse response in time of an
electronic system. The fourier transform of the impulse response gives the
transfer function (frequency and phase response). Multiplying the transfer
function times the fourier transform of the input signal gives the fourier
transform of the output signal. The main point is that convolution in the
time domain is equivalent to complex multiplication in the frequency
domain. If you multiply and phase shift appropriately in the frequency
domain, you remove the distortion, without knowing what causes it. Also, if
you don't have a computer, multiplication is easier than convolution.
It appears that they might just do a numerical optimization though rather
than think in terms of frequency filters. It's been too many decades since
I was involved with this stuff.
Hmmm ... my musical talents are nil ... maybel I should shoot black and
white.
Jeff
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|