Re: [ARSCLIST] Dynamic-frequency Range

[Aaron Z Snyder <azs@xxxxxxxxxxxx>]

Steven C. Barr wrote:

> ... ANY "improved" or "reconstructed" version of 
> that sound recording is...in fact, MUST be...based at least 
> in part on guesswork (insofar as what can't be heard in the 
> recording).
> So, the task of the program creators becomes increasing the 
> accuracy of those guesses.

There is a lot of guesswork required to "improve" or "reconstruct"
recordings, and this is one of the reasons why I have only limited faith in
programs which automatically try to do too much other than click removal and
some forms of noise removal. There are many instances of my having to find
some of the sounds, be they tones, overtones, or subharmonics, which are
buried in the noise. (It isn't inconceivable, however, that eventually it
might be possible to feed a score into a computer and have a program to
search for the inaudible sounds.) Additionally, when it comes to
compression, no automatic system can decompress recording which have been
subjected to manual gain riding by an engineer.

Steven Barr, and later Don Cox, suggest determining the transfer functions
of old recording apparati so that the resulting recordings can be
de-convolved. That's what Thomas Stockham tried to do in the 1970s,
primarily with recordings by Caruso. The results were fascinating, although
the quality was quite variable. Among the basic problems were and are the
following:

1) It is virtually impossible to get a truly accurate transfer function for
an acoustical recording environment, since there is such an incredible
number of variables, including the size of the horn, the acoustics of the
recording room itself, and the distance from and angle to the horn of the
performer(s). 

2) Add to this the fact that, even on a single-sided recording, a particular
instrument's sound quality might vary due to actual physical movement of the
performer, i.e. moving close to the horn for a solo, and then moving away to
allow other soloists to be spotlighted. 

In the end, human intervention and lots of guessing must be employed if the
overall "improved" recording is to sound consistently improved.

Aaron Z