[Table of Contents]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ARSCLIST] Dynamic-frequency Range

On 21/10/06, George Brock-Nannestad wrote:

> ----- if you read Thomas Stockham's "blind deconvolution" paper in the
> IEEE Proceedings (not to hand, so no precise reference, alas) you will
> be disappointed to find out where his logic breaks down. Essentially
> he says (my interpretation) "Caruso was a tenor, but only recorded
> acoustically. Jussi Bjoerling was a tenor, and he recorded
> electrically, so the linear distortion to the tenor voice via the
> acoustic recording process may be determined by measuring tenor tones
> on corresponding acoustic and electrical recordings. And since there
> was only one recording horn, you can then apply the inverse filtering
> to the acoustic sound."

This is completely different from what I was suggesting. You need an
impulse signal such as the sharp click from a spark to derive a function
for deconvolution.

Simply comparing one tenor with another and adjusting some EQ is not at
all the same.


Imagine an image which has been spoiled by camera movement during the
exposure (or in the case of an astronomical image, by air movement).
There will be a certain smudge which is applied equally to each pixel of
the image. The image is said to be convolved with this "smudge"

In reality there will be a complex function involving optical
aberrations, light spread in the sensor, and movement.

Provided there is some point source in the image, such as a star, which
we know should be represented by one pixel, then the function can be
measured from the actual multi-pixel image of the star. It is then
mathematically possible to deconvolve the image and remove the
aberrations and smudging (in practice, to some extent).

The nearest audio equivalent to a star is an impulse signal such as the
sharp click generated by a spark at high voltage. 

The big problem is that noise makes it hard to derive an accurate
convolution function. However, averaging from a number of clicks would

> ----- unfortunately this is a simplistic and flawed model for the
> recording process and postulating a similarity between Caruso and
> Bjoerling. As good tenors both had a peak in energy between ca. 2500
> and 3300 Hz (the "singer's formant" -- Vennard simply calls it "the
> '2800' "), and it is usually one of the signs of a good operatic
> voice. Caruso did not apply this feature, except when he needed to,
> but Bjoerling's voice had a very strong "singer's formant", and in his
> case it was a steely quality that was always turned on. Now, if you
> simply apply a first order high pass filter to the SoundStream
> (Stockham) processed Caruso records, lifting the treble (removing the
> effect of his low pass noise filter), you suddenly hear Bjoerling's
> voice with Caruso's modulation. Strange to your ears. Stockham also
> worked on John McCormack's records, and in this case at least the
> voice belonged to the same person, so perhaps the chance of getting
> there was greater.
> ----- I have used some differing approaches, one of which was
> reconstruction of acoustic disc (not cylinder, like Fesler)
> recordings, using original Gramophone Company recording gear, some of
> which was now independent of EMI for historic reasons. Most of my
> results are not yet available publicly for confidentiality reasons (a
> world class tenor, and a record company that needs to be careful paid
> for the experiments). However, a couple of years later Sean Davies
> made acoustic recordings of a lesser tenor, and those were published
> at low intensity by EMI. The original pack of materials documenting
> that also had a video showing how the recording was performed - an
> artificially "aged" documentary.
> ----- I presented some of my results (even before the above
> experiments) at the 1986 ARSC Convention in New York. I know that
> after that, some of the sound restorers active at that time started to
> use similar approaches.
> ----- finally I should say that if you correct e.g. the Melba distance
> test recordings by inverse filtering the recording horn, diaphragm,
> and stylus bar, with headphones you have the most uncanny feeling of
> her receding and then approaching, even to use a reasonably low voice
> directly into your ear. The sheer depth of that simple recording is
> fascinating when it is correctly reproduced. I have no idea if the
> commercial CD versions permit this.

I'm not sure what you mean by "filtering" but I suspect it is some kind
of frequency-dependent approach. Deconvolution is not a form of EQ.

The Melba distance tests do give a strong effect of her receding on the
Ward Marston Naxos transfers, on my equipment.

Don Cox

Don Cox

[Subject index] [Index for current month] [Table of Contents]