Hi Eero,

Yes the first thing I noticed when listening, and then looking at the orange
spectrograph in RX was how tightly defined was the whistle against the rest
of the programme. It is a very narrow, clearly defined tone. I think that's
why it's so relatively easy to deal with and why I suspect it was used as a
demonstration example.

Using high Q filtering it's been possible for a long time to filter out a
narrow tone or tones with sometimes undetectable effect on the program. The
problem with broadband noise (or swish which I take it is just a variant of
it) such as in a 78RPM record or tape is that it can easily cover the
entire spectrum of the voices and instruments in the recording. If we
attenuate the broadband noise spectrum, we have simply turned down the
volume of
everything.

For the designer of the tool there has to be some sort of ability to
"separate" the unwanted broadband noise from
the wanted program.

 Either:

1. identify the complex, sample by sample, characteristic of the random
broadband noise, so the tool can attenuate it, or

2. identify the equally complex, sample by sample, characteristic of the
music, so the tool can extract that from the noise, or

3. "replace" the content that is masked by the noise, with equivalent
content from somewhere else in the recording.

In other words there has to be enough "information" in the recording itself
for the software to do its complex discriminating work.

That's my understanding anyway.

Regards
Tim


----- Original Message -----
From: "Eero Aro" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Tuesday, March 06, 2018 5:37 PM
Subject: Re: [ARSCLIST] Swish


> Tim Gillett wrote:> My guess is the lassoo technique can
>> work with a simple patterned noise such as the audience member whistle
> ..snip..
>> but not so sure it would work on swish or hiss which is essentially
>> random
>
> To find out, you need to try it yourself. I have done it (seen no T-shirt,
> though)
> and got some of the distraction of the swishes reduced. Did it for a
> shortish period
> in the beginning of one song and my customer was happy with the result.
>
> As you point out, it is more difficult to define the area to be attenuated
> in a noise-type swish than it is with a tonal distracting sound. A whistle
> sound
> appears rather sharp edged in the spectrogram and often you can use even
> the
> RX Magic Wand selection tool for that.
>
> In a record swish the frequency content and the gain change smoothly and
> it doesn't
> have any sharp borders.
>
> That's why you need to advance with small moves, like an archaelogist with
> a soft
> brush. You draw kind of "height curves" with the lasso tool around what
> seems to be
> the most distracting area at the moment. It takes time and you need to be
> patient.
>
> I guess that such an algorithm wouldn't be impossible for the developers
> to build,
> they have done more complex things, but it is possible that the demand
> for swish
> detection in a restauration program is not large enough, so that they
> would start
> developing it.
>
> Eero
>
>
>
> -----------------------------------------
> Eero Aro
> Audio Restoration Tonfiks
> https://www.tonfiks.fi/english


---
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus