On 4/30/20 6:18 PM, David Gilbert wrote:
I confess to having generated those files, although it was several years
ago that I put them on my web site. They just get referenced here or
there about once a year when a similar topic surfaces.
http://www.ab7e.com/weak_signal/mdd.html
One dB was about all I could claim was distiguishable, but a difference
two dB was surprisingly so.
73,
Dave AB7E
this is known as the "just noticeable difference" thing and ties into
the Weber-Fechner law which is really more that perceived differences
are log scaled (that is, a 3 dB change from 3 to 6 dB is perceived as
the same jump as 10-13 dB)
for speech it's around 2-3 dB in SNR
(https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4335553/ was the first
hit in general, https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4871229/ is
another - apparently the hearing aid business really cares about this)
However, for detecting tones in noise, or constrained speech, the JND is
smaller. And, JND isn't necessarily related to intelligibility or
detectability.
I used to work with tactical comm radios just as digital coding was
coming into play and there were tons of arguments about various schemes
of various computational complexity: is CVSD better or worse than mu-law
with companding at data rate X? What about LPC-10?
It is *highly* situation dependent. What's ok in a tactical situation
where you just want to call in artillery fire to the correct grid square
is different than whispering sweet nothings to your sweetie.
This kind of analysis, of course, is why modern cellphone audio quality
is so bad. Back in the 80s, if someone had said that random consumers
would be happy with 8kbps encoding of speech they would have called them
crazy. What we used as the "standard of comparison" was so-called "toll
quality" which is about 56 kbps, sampled at 8kHz.
This is sort of interesting and describes the "signal detection"
problem, which is related to the "CW detection" problem
https://www.weizmann.ac.il/neurobiology/labs/ulanovsky/sites/neurobiology.labs.ulanovsky/files/uploads/behavneurosci_lecture13_dovsagi_psychophysics_05july2018.pdf
And of course, the real problem in CW detection is not just a random
tone in noise, it's a "structured tone sequence" in the presence of
other "structured tone sequences".
From a psychophysics standpoint it's an interesting problem
_______________________________________________
_______________________________________________
TowerTalk mailing list
TowerTalk@contesting.com
http://lists.contesting.com/mailman/listinfo/towertalk
|