Another thing:. google-search 'DSP voice recovery'. This is presumably a very old need.

https://dsp.stackexchange.com/questions/25332/voice-recovery-with-high-noise-low-gain

A voice can be modeled as a frequency, modulated in frequency somewhat, and also modulated in amplitude.

See the movie, The Conversation" with Gene Hackman.in which a filter was applied to remove some high-frequency interference.

After filtering, the man was heard to say:

"He'd kill us if he got the chance".

On Tue, Jun 29, 2021 at 12:45 AM, jim bell

<jdb10987@yahoo.com> wrote:

Look into DSP, or Digital Signal Processing.

ICs that do that haven been available since the late 1970's, but for non-real-time analysis, it should be available in software. See TI (Texas Instruments), as I recall.

You will probably want a synthetic filter whose passband is dynamically adjusted.

On Mon, Jun 28, 2021 at 11:52 PM, grarpamp
<grarpamp@gmail.com> wrote:

Since it's only Joe's annoying voice you want,
a general damp of everything left and right of it
could help too in addition to the selective cuts.
Old school analog PE applies curves from boring circuits.
New digital math is perhaps capable of much more
at the research whitepaper and secret levels,
such as shaping the curves, following a center
as it moves across the spectrum, etc.

Great Movie Scenes: Episode 9 - The Conversation