First I’m going to talk about compression – not file compression like in zip files, but audio dynamic range compression. This type of compression has nothing to do with fitting more data in less space – dynamic range compression is a technique used to bring the overall volume level of an audio track up, at the expense of volume range. In other words – throughout a track, there will be both very loud bits (the crash of a cymbal and the beginning of a chorus are normally loud parts of the track), but there will also be quieter parts – breaks and solos. Compression allows you to trade dynamic range – ie, the difference between the very quietest bits and the very loudest, for an overall increase in volume.
You can see what that looks like in these two examples – these are audio recordings of a couple of my tracks:
Not compressed
Highly compressed
Notice how both use the same range on the graph – the uncompressed signal just about touches the tops and bottoms of the graph, whereas in the compressed graph we see that the wave is regularly filling the entire range. However, even in the compressed signal the waveform never extends past the limits of the graph – you’re basically getting an increase in perceived volume without an increase in amplitude. In other words; the music sounds louder without having to actually turn the volume up. You can listen to the uncompressed sample here and the compressed one here – notice how the second sounds significantly louder than the first.
There are a lot of reasons why you’d want an increase in volume – think about music being played on the radio; the louder and punchier your music sounds; the higher the chance of the listener actually noticing it. The same concept applies in advertising. Adverts often have highly compressed audio tracks – they want to be heard and to do that they think they have to shout as loudly as possible.
It’s compression that’s to blame for the fact that when you’re watching a movie on TV and it cuts to an advert break, the adverts will often seem very loud. Films have a much wider dynamic range on their audio tracks because they actually use dynamic range to create tension – in a thriller for example, the audio track will often get very quiet just as something dramatic is about to happen, and then a sudden increase in volume will come as the shock or thrill is revealed. This wouldn’t be possible without a wide dynamic range. Film realised long ago that dynamic range is a useful artistic tool and not something to be thrown away for the sake of shouting louder. The only drawback to a wide dynamic range can be that if you’re trying to watch a film quietly, you find yourself constantly adjusting the volume as the film goes between whisper-quiet scenes and shouting-loud ones.
Before I go on to explain why I think compression is killing music, let me first explain that it’s not always bad. Most electonic dance music uses huge great bucket-fulls of compression on the drum track – so much so that it actually distorts. This deliberate distortion with a compressor of a driving repetitive drum line originally produced by the TR-808 or the TR-909 (or both) was what kick-started electronic dance music. These machines coupled with a lot of compression are capable of creating the stomping dancefloor beats that pilled-up teenagers have been pounding dancefloors-to ever since. Compression is a key ingredient needed for electronic dance music and I’m not suggesting that we should do away with it entirely.
However, the trend has been for more and more compression to be applied during the mastering stage of the music production process. The net result? Louder and louder tracks on CDs, on the radio and on pretty much every other distribution mechanism. Nobody wants to be the one who releases a track that sounds significantly quieter than everybody else’s. What I’m guessing happens is that the tracks come from the recording artists and the producers already with plenty of compression on each track, then the mastering engineer goes and slaps a load more across the whole mix to bring the loudness up to levels that are acceptable within the specific genre.
It does vary quite a lot across genres – pop music is always very loud these days, as is pretty much anything with a guitar in it, but classical music is still recorded with a much wider dynamic range. Different genres of electronic music vary quite a bit in the expected levels of loudness – drum and bass, house and trance all tend to be very loud, but downtempo, dub etc get away with lower levels.
The problem with overcompressing a track, especially when the compression is applied to a mixed-down version of it is that it can lead to noticable ‘breathing’ – at its worst, it can sound like somebody is literally sitting there and turning the volume up and down in time to the beat. Perhaps to the untrained ear this is bearable, but I recognise it as an artifact of compression and it makes music that I would otherwise quite like completely unlistenable. A certain amount of compression is a good thing – it makes a mix more punchy and in the case of dance music, gives it the drive to get a dancefloor moving. It’s when compression is overdone just for the sake of increasing levels that it starts to get a bit annoying.
The other annoyance of the tendancy towards increased loudness is that it robs music producers of some of their freedom to play with dynamic range in the same way that film producers can. When all music has to be at a certain loudness, it doesn’t leave much room for the whisper-quiet bits like you get in films. It’s a case of money-grabbing record companies trying to make their product more marketable at the expense of actual creativity.
Dynamic range applies to photography too – when talk about the dynamic range of an image the easiest way of looking at it is a histogram. Here’s an example of a good histogram:
Coastguard cottages on the cliff
Good Histogram
The histogram shows the tonal balance of the image, basically a lightness value is calculated for each pixel in the image, and the number of times each lightness occurs is counted. You can think of it as desaturating the image (making it black and white) and then counting how many pixels are each shade of gray. The histogram displays the counts for each level starting with black on the left and ending in white on the right.
What we’re normally aiming for is a good range of tones starting right at pure black and extending throughout the tonal range all the way to pure white. You can see in the histogram above that we’ve pretty much covered that – there’s detail in the shadows which accounts for the range of tones in the first segment on the left – then most of the image is taken up by the sea and the hills which account for the peak in the second section. However, there’s also lots of details in the clouds which are much lighter, and this accounts for the second main peak in the fourth section.
If the image is underexposed, it’s too dark and the levels on the histogram will bunch towards the left. If it’s overexposed, it’s too light and the histogram will be bunched towards the right. Have a look at these examples:
 Underexposed Hay Bail  Underexposed histogram |
 Overexposed hay bail  Overexposed histogram |
Armed with the knowledge that you want a wide tonal range, it’s easy to take photos which do not cover the full range and ‘correct’ them to make them look better:
 Original from camera  Original levels |
 Corrected version  Corrected levels |
What we’ve actually done here is increase the dynamic range of the image by essentially turning up the contrast. You can see the effect it has on the histogram – it’s stretched so that it covers a greater tonal range. The photograph at the time was taken with the optimum exposure bias to get the best tonal range – I check this while I’m shooting by looking at the histogram on my camera’s display, but I also take a number of exposures (bracket) at different biases so I’ll always end up with a range of images from slightly underexposed to slightly overexposed. With my landscape photography I always pick the exposure that I feel most accurately represents what I saw with my eye and I don’t apply any corrections on it.
You might agree that the corrected image above is ‘better’ – it’s more vibrant certainly and you can see more detail in the shadow. You can adjust the dynamic range of an image quite accurately in a number of ways with a modern graphics package like Photoshop or The GIMP. Most of the time the only tool you need to use is the ‘Levels’ editor.
So why don’t I apply these corrections to my images? Well, there’s a number of reasons. Let me first retract what I previously said about ‘Good’ histograms -that you normally want the full range of tones isn’t necessarily true. There is no such thing as a ‘Good’ histogram – only a histogram that matches the photograph that produced it. It is the photographer’s artistic choice whether they use the full range of tones or not. I may choose not to. You wouldn’t say a painting was wrong because it’s made mostly of black and dark coloured paints – it may just be a painting of something dark, for a photograph to only contain a limited range of tones is not wrong in itself, it’s just not normally what you want.
This is partly born out of the fact that over and under-exposed images will tend to bunch towards one side of the histogram; it leads people towards the false assumption that a histogram that’s bunched together is always bad or wrong. It may not be. Imagine I have a piece of card that is a middle shade of grey, perfectly even, perfectly lit, and I have a perfect camera – I want to make the most accurate representation of that grey piece of card that I can.
Grey card
Grey Histogram
As you can see, this image has no tonal range at all – there is simply one peak on the graph, on middle grey. It is however an accurate representation of our perfectly grey piece of card. What I’m saying is that there is no link between tonal range and whether the photograph is an accurate representation of reality. A scene simply may not contain the full range of tones, at night it often doesn’t. Furthermore, even if you’re not trying to accurately represent reality; sometimes you just don’t want the entire tonal range:
Headlights at night
Night Histogram
In this image produced in the chemical darkroom I’ve deliberately increased the contrast to the point where every pixel is almost completely black or completely white. That was the effect I wanted – the image wouldn’t look right any other way. I think that proves that there’s no such thing as a ‘wrong’ or ‘bad’ histogram.
I have been known to manipulate the dynamic range of an image – I did it all the time in the chemical darkroom and the image above is an extreme example of that. So why did I stop when I switched to digital? Well, because digital allows me to take so many more images, I can do exposure bracketing, take several exposures and then pick the best one – the one that I feel most accurately represents what I saw – this doesn’t always mean the image with the widest dynamic range.
A lot of the reason for needing to do corrections in the chemical darkroom was because your original negatives weren’t perfect exposures – digital allows you to get closer to the ideal exposure simply because you get more shots at it because you’re not paying for film. It makes a lot of the corrections unnecessary because you simply pick the best image and don’t fiddle with it. When you can get away with using the image directly as it comes from the camera it makes a lot of sense to do so.
Because digital produces so many more images, it’s no longer practical to manually correct each one anymore and I wouldn’t want to even if I could. To unnecessarily adjust an image to make it look better doesn’t fit with my goal of accurately capturing what I see; even if I know there’s an adjustment I can make that would make an image look better, I wouldn’t do it unless I felt it necessary to help the image be an accurate representation of reality.
Photographers who always adjust their images so that the histogram fills the entire tonal range often end up producing images that to me look lifeless. In some ways adjusting images to fill the tonal range is similar to applying compression to an audio track – it makes it more vibrant and punchy, but at the expense of some of the subtlty. It’s okay to a certain degree, but it is possible to over-do it and end up with images that just don’t look real anymore. Unfortunately, like with audio compression, the tendancy is towards all images being manipulated in this way to an increasing degree. When everybody’s images are enhanced in this way, you don’t want to be the odd one out whose images look dull in comparison.
However, it’s important to me to capture what I see, when you start manipulating an image it becomes what you wanted to see rather than what was actually there. Besides, as soon as you start playing with the contrast and colour levels in an image, you forget what you originally saw and the goal simply becomes to make the image look as good as possible.
It doesn’t fit into my workflow more than anything – I want to have the original JPEGs straight from the camera uploaded to my gallery which means any levels corrections would have to be done via a web browser – this is a feature Coppermine doesn’t have, but I will almost certainly add it at some stage. But even when I do have the capability to perform levels corrections on my images, that’ll be minor alterations in a few cases just to brighten up one or two important images, I won’t be applying corrections to every image that I upload. I just don’t believe it’s worthwhile to manipulate every image so that it fills the full dynamic range – I don’t even agree that filling the dynamic range is always the goal.
Let’s see another example of expanding the dynamic range:
 Hay bail - original from camera  Haybail Original Histogram |
 Hay Bails - Corrected  Hay Bail Corrected Histogram |
