bbystrek wrote:
>
> Hello,
>
> Reading some of the back posts regarding using an omnidirectional
> microphone with a parabolic setup in an attempt to cancel ambient
> noise got me wondering...
>
> The omni was either hard wired in an antiphase fashion or recorded as
> a second channel and later applied during post recording editing.
> >From what I gather, everyone seems to have abandoned the experiments
> due to an increased perceived noise floor speculated to be due to
> spatial "time of arrival" (phase) issues.
>
> Could anything be gained from collecting a noise sample to later be
> used to feed the noise reduction algorithms incorporated into many of
> our digital audio workstations? Perhaps either recorded using an
> omni, or simply pointing the dish at say the barking dog in the
> distance for a few seconds. If nothing else, I bet it would help
> guide any digital manipulations later; or at least provide critical
> information as to what spectral components are present in each noise
> source. Though I suspect that a lot of the unwanted noises we record
> are reverberant in nature, modified by both the environment and the
> parabola characteristics. Directly pointing a parabola at the noise
> source would not necessarily represent the spectral content of the
> reverberations as they appear in the recording of our subjects.
Adaptive noise filters work off a very short sample from the original
sound. Moving the mic or pointing it different will change the sample so
much it would be useless on the original the filter is working in a much
more detailed way than that. At least with the noise filter I use I
simply manually choose a segment to sample, typically grabbing between
calls. The default time of the sample in Spark XL is 500ms, though you
can choose over a wide range.
Note this is a place to talk about fiddling with the gain during
recording. If you do that, just what part of the track will you sample
with the noise filter to calibrate it? It's a big reason for the set it
and leave it school of record gain setting. Even then the character of
the noise will vary some over the length of the recording even if it's
just a few minutes. But you have a chance of getting good results from
noise filtering.
> Anyway, has anyone tried anything along these lines? Perhaps some of
> the experts amongst the software manufacturers would be willing to
> discuss how to optimize their algorithms.
I optimize the noise filter by listening and by watching a realtime
sonogram. It varies with each track I do, and also varies depending on
any changes to filters that come before the noise filter. I've found
that most of the time Spark XL's 500ms timing is close to optimum. It's
also important to not overdo any of the settings, my filter has several
different settings to it the most obvious being the dB level, which I
rarely set more than -6 or -7 dB, and often less.
Barking dogs are not removed with a noise filter, that's used for steady
noise in the same range as the calls you are interested in. Sometimes
it's possible to use frequency cutoffs to minimize a dog barking. It
depends on what frequency the calls of interest are.
In my case I'm almost always recording stereo with the Telinga, so
adding a noise reduction mic would be channel 3 (and maybe 4).
Walt
________________________________________________________________________
________________________________________________________________________
|