Freesound Forums

So you want to record nature, but are not happy with the noise ambiance you get? Here are a few tips and advices from my recent practive. I don't know whether are they "professional" or not, but they seem to work pretty well. Perhaps this little text will encourage some of you to share your thoughts about.

What you are going to need is:

1. multitrack software for mixing at least 2 tracks. There are a lot of programs on the internet, some free, some cheap, and some expensive.
2. You need some EQs, noise reduction plugins/software (which cleans the tracks, even if the results sounds a little bit artifacted like cheap and wet mp3) and noise shaping plugins/software.

Okay.

On your recordings, there are probably some unwanted sounds, coming from wind or far away cars, airplanes and ambiances like that. Many of them - operate in low frequency specttrum, which means - you can take their power even if you cant remove them completely.

If you set the hi-pass filter to "Q=7" or something around (level of resonance, which means - how sharp should be the filtering curve; too sharp is not good because it sounds like "howling wind" or a box, too soft will take too much from your sound or too less, depending on gain), it will be fine. It's a gentle setting, from what I see - it is a default value in dx/vst plugins. Filtering bottoms means, that the threshold frequency is the beginning point of the process, not the filtered end (so you will have lower freuencies, but with less power). On electronics they taught us, that the threshold frequencies in pass filters are used to mark -3dB (or c.a. 70%) dropdown of reference signal. 6dB is 2x lower loudness.

Decibels are strange animals, because they are relative units. Something in relation to something else. Therefore - having two mp3 players, with different maximum loudnesses - both of them have "zero dB" as "maximum" reference. And everything below these references - is relative to these individual "zeros". -6dB is like half of 100% (=50% total loudness), -12dB is like half of the half of 100% (=25% of total loudness) and so on. Also - although decibels used for your mp3 players are the same units like these used for measuring sound pressure, they represent different meaning. For example - let say that your microphone can handle 100dB. But decibels of microphone represent different kind of relativity. 100dB for microphone means, that the microphone will probably get sick if you try to record something like loud percussion or busy airport in close view (and no matter how you set your decibels on your recorder - which are in relation to "maximum zero" of the recorder - the sound will be overdrived let say at minus 12dB). Decibels are strange animals.

You will need some low frequencies if on your recordings there is not too much going on or - to get warmer / deeper / more full bottom. Like subwoofers do. But I guess you can remove all below 65Hz. Usually you can remove all below 100Hz, but depending on sonic content - you can reach higher levels - up to 300Hz (the higher you go, the more mobile phone / multimedia speaker it will sound; by the way - multimedia speakers usually filter everything below 100Hz up to 200Hz). I'm using thresholds like 65Hz, 100Hz, 150-180Hz, 250-300Hz to remove cars and some other industrial sounds. Depending on what you record - birds or frogs or other ambiances - check the optimum values to get the most natural reaults you can get. Removing lows enables you to get higher gain possibilities, because most of what you remove - is hi-gain non-preceivable wind pressure (especially if you don't have a good hairy windjammer, but only spunge).

There is one interesting thing to mention here. Google or search Wiki the "hearing curve" to understand it. In different frequency bands (pitches) - sensitivity of our hearing is different. It reacts differently to gain (amplitude. decibels) changes. For example +12 decibels will give you more noise and lower frequencies, but very similar subjective loudness of hight frequencies like from birds. So you don't need to normalize bird recordings too much, because they will sound similarly loud, but more noisy if you take higher values. By the way - for birds only, mixed with other ambiances - you can setup the highpass filter even around 1500-2000Hz. It will remove a lot of other sounds, and birds will be untouched.

Also - you could play with EQ bandpass (at Q=7) to make higher or lower thresholds (+/- 3-12dB probably) somewhere between 300-900Hz; it will change the colour of soundspace.

Now to some practice.

Make one "raw" recording, maybe with basic filtering (like wind filtering at 100Hz) but no dynamic / aggressive effects. Make another recording, adding so much noise reduction and other effects effects as you wish, a little bit more than you think that it is okay. The second recording will probably sound like too much filtered, a little artifacted (affected by noise reduction plugins) and wet, not natural.

Start your multitrack mixer/software and synchronize both tracks 9rav and noisereducted). Now you can manipulate how much of original, raw sound you want to have, and how much cleaned. Your raw recording will be probably somewhere at -6 to -18dB in relation to noisereducted part. Lower the gain of your whole project (2 to 6dB will be probably enough), to avoid gaps; you can normalize it later. Listen to it, to find optimum values (somewhere between raw and NR version) And then - mix it, perhaps adding some noise shaping algorhithm, to reduce the wetness. The result - you get clean but sharp, natural sound. Mixing itself - does make change, because the software is always trying to determine best values between subparts (like antialias in graphics).

What else you can do? On your mix, you can find uncertain areas, in which you can play with gain of both versions (raw and NR), to gently remove certain specific/dynamic artifacts.

And now - some sleep.
(-;

Update.

Let say, that you have recorded something really really wonderful. Many interesting sonic events going on around, many voices, many different sounds. But... in the bacground you hear however, pretty often - these annoying sounds like dogs. I have nothing against dogs, I like them. But frustrated furious barking is annoying and can destroy a lot of peaceful work. To cut out dog scenes - impossible, too many scenarios will be affected. To just filter with bandpass EQ - impossible, because it will lose its natural sounding.

While recording bees in blackthorn, and getting many interesting bird talks, my barking dogs from distant rural areas - operated somewhere between 600 to 900Hz. Cutting exactly this part of spectra (via accurate FFT filter) - removed all dogs, but box-like sounding effect appeared. Is it possible to make up such thing?

Well - actually yes. Many soinic events operate in their specific frequency spectras, and everything else - is like homogeneous tonal noise pattern. Bees are rather homogeneous in nature (they just make "bzzzzz" sound, changing its shape), birds are in much higher frequencies.

So here is what you can do. First recording is raw, no freq removal. Second recording is band filtered between 600 and 900Hz (sharp edge / high Q). Use them side by side in multitrack, exchanding non-barking-raw parts with barking-filtered parts. Crossfade should be linear, and somewhere from 5 to 10 seconds long, to make it natural. And then...

...if on your raw recording - you have clean sections, with no barking dogs - copy them, connect with each other to make a longer time-line, and filter them with bandpass filter, removing everything except this 600-900Hz band. The same bandpass filter you have used but in opposite direction, to clean your raw recording from barking spectra.

Then you can use this third file, to mix (adjust ypur decibels, depending on initial filtering depth) with filtered version of your raw recording. This small 300Hz freq band - will fill/suplement your recording with missing tonal noise pattern. And since there is nothing special going on in this band - the result will sound very naturally.

You can act in similar ways with different sonic spectras, depending on how wide they are, and what is inside them. It will not work if there are sonic events, although - you could find similar events, to fix it.

*

Trust me - if you give such recording to someone else - they will not know nor notice your work. It's like putting a small stick in the water, while brook itself - is very busy and dynamic anyway.

GREAT INFO

i always find that Soundsoap by Bias is also an awesome and easy way to pinpoint and take out certain noise frequencies in post while being able to preserve all the other frequencies you want to keep .. and, no I don't work for Bias or anything, it's just a fantastic product

You just made my day, ayamahambho. I've been making long "nature" recordings for a couple years now and finally getting to where I want to make them into a collection, and I can't wait to try your techniques.

ayamahambho
And then - mix it, perhaps adding some noise shaping algorhithm, to reduce the wetness.

Great info Ayamahambho! I use this technique for mixing various effects but it never occurred to me to use this for cleaning work too. Wow!

Regarding the above noise shaping algorithm; are you referring to a frequency attenuator / hiss removal or more in the lines of a surgical EQ? I'm interested in trying these out.

Much thanks for your thoughts and time putting this together!

Noise shaping as a part of dithering.

To make an analogy - in graphic apps it's similar to sharpening or softening edges and textures. Try to resize a picture made of lines and edges, and use different resizing algorhithms. Seeing these results will give you the general overview or understanding of - how noise shaping works on audio files.

The "perhaps" part refers to the fact, that you don't need to change these parameters always; default can be good too.

Here are some of my nature snapshots:

http://conscious-sound.bandcamp.com/album/binaur-e-al-sonosphere

Auditions are free to listen online. All recorded in Poland, 2011.

p.s.: Bandcamp offers lossless FLAC downloads.
Online listening is MP3 at 128kbps according to their specs.