De opgenomen geluiden bewerken / editing recordings

In Nederland zijn de opgenomen geluiden zelden vrij van storende bijgeluiden. Daarom moeten de geluiden meestal bewerkt worden. Dat gebeurt op de computer met speciale software. Een veelgebruikt software-pakket is CoolEdit. Onderstaande bijdrage van Hans Dankbaar behandelt de meest voorkomende bewerkingen, tips en trucs.

In our country most recordings have unwanted sounds in the background. It is therefore necessary to use sound-editing software. In the contribution below Hans Dankbaar points out what possibilities there are when using the much-used program CoolEdit.

Sound Editing with CoolEdit 2000
            by Hans Dankbaar

The following presents a concise guide for sound editing as can be done with program CoolEdit2000, including many hints and tips.

Introduction
By editing sound recordings, you are going to alter the original recordings. Before you start, it is good to consider what you want to achieve with it, how far you want to go with editing, and which (edited, original, intermediate) versions you want to keep and in which data formats (e.g. wav or mp3). Because I mainly record bird sounds, for me the most important reason for editing sound is the removal of all kinds of noise and I want to achieve bird sound tracks with as little noise as possible.
It needs no say that one should avoid recording noise in the first place, through good quality equipment, recordings from as close a distance as possible, and in a low-noise situation. However, close proximity is difficult to achieve with birds, and noise free places or moments in time are often scarce: common sources of noise are weather (rain, wind), traffic (air-planes, cars, etc), voices from and noise generated by nearby people and microphone-contact. Sometimes also loud calls from other birds can be perceived as noise. 
With a dedicated editing effort, you will mostly be able to distil a reasonable quality sound track from an apparently poor (noisy) recording. Objective of sound editing should be to remove noise while not affecting the desired sound. This implies that critically listening and judging the results during the editing is very important (use e.g. a headphone). Also compare your final result once more with the ‘raw’ original recording.

CoolEdit2000
This program was for sale in 2003 for about $75 from a company called Syntrillium; there also existed a ‘Pro’ version for a considerably higher price. Since 2004 the company was bought by Adobe, who did not continue the support of the ‘normal’ CoolEdit2000 version, but only the ‘Pro’ version, for a price of about $300.
CoolEdit2000 is however still used widely by sound recording enthusiasts. It is a very user-friendly program with many and also advanced editing options. In fact it has almost everything you would want - and even more than that.

CoolEdit – program window:

Centrally located in the window is the image of the sound. There are two types of images possible: the ‘time domain’ display and the ‘frequency domain’ display. In the frequency domain display you see the strength of the sound (indicated by colour, where dark=weak signal, yellow=strong signal) as a function of frequency (vertically) and recording-time (horizontally) – see fig 1; in the time-domain display you see the air movement as function of recording time (like a guitar string – see fig 2).
For most purposes one needs the frequency domain display, but for certain applications (e.g. removing tics) the time-domain display may be more appropriate.

Immediately above the sound image one finds the buttons for fast access to the editing functions. One can add or remove additional fast-access buttons here, as explained below (‘Hints & Tips’).
 Except for these buttons, editing functions can also be accessed     through the function-keys from your computer keyboard (e.g. F1, possibly combined with the Ctrl and Alt keys).
The access through the keyboard function keys can also be configured, as explained below (‘Hints & Tips’). 
At the very top of the window (File Edit View ...) one finds the standard access to the functions, options and settings.
In the bottom left, the buttons for play/stop/record are located, immediately to the right of those are the buttons for zooming; in the centre is displayed the time at the cursor position (millisecond accuracy), and again to the right the times of the (zoom) selection.
Some important function buttons are shown in fig.2, like the ‘undo’ button, and the button to switch between time and frequency domain display. 

HINTS & TIPS:
- Start/stop playing: use spacebar from keyboard.
- Display time and frequency at cursor position: RM (right-mouse click) on the very lowest edge of the window (near ellipse in fig.2), and in the pop-up window that appears, select ‘data under cursor’. Also filesize and other information can be selected to be displayed in the bottom-row of the window. 
- Zoom vertically (frequency): use button with magnification glass at lower right hand corner to zoom in or out; the zoom centres at the middle frequency; then scroll up/down to higher/lower frequencies by pressing MB1 on the frequency-axis, and while keeping MB1 down, shift cursor up/down.
- Scrolling horizontally: same as scrolling vertically, but press MB1 on the horizontal (time) axis.
- Fast-access to the functions: this can be achieved in two ways: 
1. Add buttons for these functions (toolbars) or 2. define function keys on the keyboard (shortcut keys). Both these fast access methods can be configured by clicking ‘Options’ on the very top-row of the window, and then selecting the desired method from the drop-down menu. With the shortcut-keys one can define almost every function, also e.g. zooming in or out. In this way one can reduce the usage of the mouse. A number of shortcut keys is already pre-defined, like Ctrl-X,Ctrl-C,Ctrl-V for cut,copy and paste, and Ctrl-‘arrow left/right’ for zooming.
- Resolution of the frequency display: the default setting of the resolution is far from optimal. Select Options-->Settings; in the menu that appears, select tab ‘Spectral’; there, right from Resolution, you find ‘64 Bands’ (the default). Here, select 128, or more, to get a sharper image. The displaying of the sound image might become a bit slow if you select very high values. 

Sound editing - general

The picture to the right shows a recording with different kinds of noise. 

Sound editing in general can be divided in three categories: 
1. Cutting, pasting and mixing
2. Volume adjustment
3. Filtering
In each of these categories one can apply simple or advanced editing methods, using CoolEdit.

Brief overview of editing methods

- Cutting: If the noise is separated from the signal by time, one can remove it by cutting it out. Disadvantage: the recording is shortened by it, which may be undesirable e.g. for the interval between two bird calls. This method is particularly suited for short duration noise (tics). 
- Volume adjustment: if signal and noise are separated by time, instead of cutting the noise also reducing its strength (reducing the volume) can be applied. Disadvantage: abrupt changes in the volume can be audible and unpleasant, unless the transitions are selected very precisely and carefully. This method is suited for longer, relatively quiet intervals in the recording which contain disturbing noise. Volume adjustment can be done manually as well as automated.
- Filtering: if signal and noise coincide in time, we have to resort to filtering. 
‘Conventional’ frequency filtering can be applied if signal and noise coincide in time but have separated frequency ranges. The frequency-range of the noise can be ‘suppressed’ by any desired (adjustable) amount.
If signal and noise overlap in time as well as in frequency, we still have a last-resort: the NoiseReduction (NR) filtering. NR can only be applied for mild suppression of noise that has a constant character with time. A too strong application of NR (or if the noise varies too much with time) can result in disturbing distortions in the filtered sound. 

Important: don’t hesitate trying things out, until you’re satisfied - with the undo button you can always undo editing steps, even up to 5 or more steps back (this number can also be adjusted, through Options-Settings). 

Extended discussion of editing methods

- Cut, Copy, Paste, Mix: (shortcut keys: Ctrl-X, Ctrl-C en Ctrl-V)
Cutting out pieces of noise/spikes: select the starting and end points of the noise accurately, especially in case the proper sound and the noise are barely separated. If needed, zoom in. Sometimes, e.g. in case of high-volume passages, cutting of noise spikes doesn’t work well: the spike remains whatever you do. In that case the spike is caused by a discontinuity in the sound signal, which can only be made visible through a time-domain display. In this domain, after strongly zooming in to the spike, the discontinuity will become apparent (see figure above) – then make a very precise cut into the signal such that it becomes smoothly continuous again. 
By cutting, we shorten the signal. In some cases this can be undesirable, e.g. if the interval between a succession of bird calls is shortened. A solution in that case is to replace the cut noise part by a piece of (noisefree) background sound, taken from nearby (e.g. from in between nabouring calls). This replacement piece should have exactly the same length as the cut part and should resemble as much as possible the cut part (but then without the noise). In the same way you could even replace a little piece of a call that contains a noise-spike by a piece of call that doesn’t contain a spike. This requires very precise editing, and the question also is how far you want to go.
Mixing: you can mix a piece of recording with another piece of recording (overlapping), e.g. to replace the background signal or to add something to the background (e.g. to mask a disturbing soft background noise). The ‘added’ signal could be tailor-made frequency filtered prior to adding it to the main recording, e.g. if you only want to add signal in a narrow frequency band.
Mixing can also be used as insertion with tapered transitions (crossfading), to make smooth transitions between different recordings.

- Volume control:
The volume can be controlled in several ways:
1. the selected part of the signal is entirely amplified with a constant factor: through functions Amplify and Normalise: Amplify multiplies with a given percentage or dB amount, Normalise derives the multiplication factor from the strongest part of the signal, such that that always gets the same strength.
2. Fade-in/fade-out: Amplification with a factor that varies linearly over the selected interval from a given start-value to a given end-value.
3. Envelope: Amplification over the selected interval by a factor that varies with time according to a user-defined function; method of adjusting this function: see ‘designing the frequency (FFT) filter’.
4. Automated strengthening/weakening of the signal by Dynamic Range Processing (DRP): here signals with a strength above a certain threshold will be strengthened and below the threshold will be weakened. The threshold and amount of strengthening/weakening can be adjusted. This method could be applied if signal and noise are separated in time but also separated by volume, e.g. to automatically lower the volume in silent passages. The adjustment of DRP can best be done graphically (Graphic-tab); the transition time (release time) can be adjusted below the Attack/Release tab. Try short release times, e.g. 5 ms. One can even select the frequency-range to which the volume-adjustment should be applied (band-limiting tab) – if you do that, then you are actually applying frequency filtering. DRP should be used for mild adjustments. For me DRP rarely gives satisfactory results, it requires that the signal is always stronger than the noise - if a bird call has soft parts (or an echo), these will also be weakened.

- Filtering
Through frequency filtering parts of the frequency range can be suppressed (noise) or enhanced (signal).
Frequency-filtering can be done with ‘Quick-filter’ and with ‘FFT filter’. 
Quick filter (QF) has the advantage that it can be adjusted quickly with shift-buttons, but does not allow very precise selection of the frequency ranges to be filtered. 
FFT filter (FFT) is (for me) the real workhorse for the sound editing. The shape of the frequency filter can be designed in any shape through a graphical interface. 
Filtering tactics:
Especially strong noise can often not be removed in a satisfactory way in one step: best is to apply a series of filtering steps, whereby each time the (FFT or QF) filter is adjusted (strength & frequency range/shape) to attack the remaining noise. A too weak suppression is harmless: just filter again, as many times as needed. A too strong (suppression) filter - resulting in (too) dark bands in the filtered signal - cannot be ‘cured’ other than by an ‘Undo’.
Designing the FFT frequency filter:
Filter corner-points:
Like with cutting out noise in the time domain, cutting noise in the frequency domain requires careful selection of the frequency start and end-points (corner-points) of the noise to be removed - do not remove more than necessary, remove ‘disturbing’ noise only; for this purpose, make sure to use the frequency-domain display (like fig.1) and move the cursor successively to the positions that mark the highest and lowest frequency bounds that you want to filter. While doing that, read off the frequencies that belong to these positions in the centre-bottom of the Cooledit window (see hints&tips on how to get the frequency under the cursor displayed; see also there on how to zoom vertically). 
Amount of suppression/enhancement:
The maximum suppression (or enhancement) can be typed in the ‘Max’, ‘Min’ boxes (lower-right hand corner in the figure), with values expressed in dB (select ‘Logarithmic’ in the upper-left). Positive values imply signal will be enhanced, negative values imply signal suppression. Select the amount of suppression (the Min value) carefully: again, do not suppress to a level that is below the ambient background noise – this can be judged by the colour of the filtered signal: the colour should not be much darker than that of the remaining background noise outside the filtered frequency range. 
Filter shape & slope:
Preferably create smooth filter shapes: switch on the tick-mark at ‘Spline Curves’ in centre. The filter can be shaped by: adding filter points (click anywhere in the window), shifting points (click on the point, hold MB1 down whilst moving the cursor), and removing points (proceed same as with shifting, but then move cursor swiftly out of the window). The steepness of the filter (roll-off slope) at the corner points should also be chosen carefully: preferably they should be the same as the roll-off slope of the noise itself: in that case there will be no remnants of the noise left; in case signal and noise are barely separated, the filter slope must be made steep. If needed, the filter can be made very narrow and steep in order to filter (suppress/enhance) a very selective frequency range. See example-filter above.
Time-varying filtering:
Both filters (QF and FFT) can be applied in a time-varying way: QF in a linearly varying way between an ‘initial’ filter (at time 0) and a ‘final’ filter (at max-time), FFT through a arbitrary variation with time between two given filters. To achieve this, with QF remove tick-mark at ‘Lock to these settings only’ and design the two filters under the ‘Initial Settings’ and ‘Final Settings’ tabs. With FFT, remove tick-mark at ‘Lock to Constant Filter’, hit ‘Transition Curve’ button to design variation with time, and ‘View Initial Filter’ and ‘View Final’ at top-right to design the filters.
Saving filters:
With both methods (QF and FFT), the designed filters can be saved by adding them to the preset filter-list: hit the ‘Add’ button at the top of the little PRESETS subwindow, and type the name you want to give to it in the pop-up window.

- NoiseReduction (NR)
NoiseReduction (NR) can be used if signal and noise overlap (coincide) both in the time domain and in the frequency domain. NR can only be used for mild suppression of noise, and the noise must have a constant character with time. The application takes place in two steps: First a piece of ‘clean’ noise (noise profile) must be given to NR – based on this profile the program itself designs the NR filter. The next step is then to apply the filter. The noise profile given should be at least 1 sec long and contain nothing but noise. If the recording does not contain a continuous section of noise longer than 1 sec, you can still create this by copying a few scattered patches of noise and paste them next to each other, such that the noise can be derived from it (undo the copy&paste afterward). The complete procedure is mentioned in the figure below.

Procedure: 
Step 1-4 describe the derivation and saving of the noise profile, ending with closing of the NR application. Steps 5-7 describe the application of the filter. Important here is the specification of the filter strength (‘Reduce by’ box). Here, 10 (dB) would already be a quite strong suppression, 5-6 dB is moderate and 2-3 dB is mild. A too strong suppression will result in disturbing deformations (often in the higher frequencies) – then undo the filter and re-apply with a milder suppression. A non-optimal NR filter can not only result in deformations of the filtered signal, but also in (just) a volume-reduction of the signal. Therefore, carefully check the volume of the signal before and after filtering. Sometimes a non-optimal filter can be made ‘optimal’ again by shifting the ‘Noise Reduction Level’ button down from 100% to e.g. 50%. If that doesn’t work, re-derive the filter from a different section of noise.

Good luck !
28 March 2006