Bit-rate compression and dynamic range compression

[steve_1979]

They're totally separate things which people often confuse as being the same thing on forums.

'Compressed' files like MP3's and AAC's use reduced bit-rates and clever algorithms to make the file sizes smaller but this doesn't effect their dynamic range which is totally separate and is unrelated to the type of file used or the size of the file.

With regards to audio, dynamic range means the difference in height (volume) between the biggest (loudest) wave peaks and the smallest (quietest) wave peaks. Converting a lossless file to MP3/AAC doesn't alter the dynamic range.

 

[Deleted member 108165]

Yeah and don't we know it on the recent crop of pop cd's, and some dodgy so called re-masters! Some of them are awful! So after 34 years without one I invested in a turntable and currently have one album to play on it.

Why can't I get the damned animated smilies to work on this forum anymore!!?

 

[Waxy]

..you wouldn't want to provoke some sort of argument involving entrenched, polarised views.

Therein lies madness! *crazy*

 

[steve_1979]

..you wouldn't want to provoke some sort of argument involving entrenched, polarised views.

No not at all, though I can see how it might look that way. Those threads bore me and I rarely bother to get involved in them these days.

I just wanted to comment on a relatively common misconception that I keep seeing on hifi forums. Some people don't seem to realise that bit-rate compression and dynamic range compression are two totally separate and unrelated things.

 

[steve_1979]

Therein lies madness! *crazy*

Been there. Done that.

 

[nopiano]

Hi Steve, your distinction is a well made point. I wonder if some of the confusion is because modern mp3 downloads are in the majority carrying dynamic-range squashed pop music, whereas (snob alert) those of us who listen more to classical stuff may be more likely to fall for hires recordings. Of course we may be deluding ourselves on that score!

 

[CnoEvil]

Of course we may be deluding ourselves on that score!

Nice Pun. *biggrin*

 

[Native_bon]

They're totally separate things which people often confuse as being the same thing on forums.

'Compressed' files like MP3's and AAC's use reduced bit-rates and clever algorithms to make the file sizes smaller but this doesn't effect their dynamic range which is totally separate and is unrelated to the type of file used or the size of the file.

With regards to audio, dynamic range means the difference in height (volume) b etween the biggest (loudest) wave peaks and the smallest (quietest) wave peaks. Converting a lossless file to MP3/AAC doesn't alter the dynamic range.

Well most who work in a music studio enviroment know this fact. Facts dnt count for anything nowadays. Just marketing.

 

[Tannoyed]

I suppose I was responsible for this thread by making a chance remark re dynamic range!

I acknowledge that compression of dynamic range (reduction of difference between loud bits and quiet bits) is not the same as bit compression but I suggest that the one leads to the other.

I may be out of date a bit but my understanding is that a digital number (1100101010101101 etc) is ascribed to each one of a sequence of analogue levels. These numbers ideally need to be evenly spread over the full range of possible analogue levels but with a typical signal, the variation is not that great, only hitting high levels infrequently with for example, a drum beat. So there are a lot of wasted numbers if you try to look across the full range all of the time. By compressing (dynamic range) the amount of variation you can use the numbers more efficiently over a smaller range. No doubt further digital mechanisms can be used to compress the data further and indeed clever things can be done these days but fundamentally the spread of numerical values is finite and to make efficient use of them when sampling, some sort of dynamic range compression must surely be carried out before sampling takes place

MP3 files achieve a lot with very little, to make use of the limited storage space available but compression is indeed what I hear when I listen to them, perhaps not too badly in its highest quality forms, but all too evident at the lower end. Every drum beat hammers the overall signal level down and then it rises again only to be crushed once more. It is most noticeable in orchestral pieces or when the beat is relatively slow. For portable devices this is fine but for hifi it is really no good at all.

Ok I am prepared to be shot down in flames now. If I have oversimplified things (or indeed got this totally wrong) I should be interested to learn more.

 

[The_Lhc]

I suppose I was responsible for this thread by making a chance remark re dynamic range!

I acknowledge that compression of dynamic range (reduction of difference between loud bits and quiet bits) is not the same as bit compression but I suggest that the one leads to the other.

If that was the case how do you account for the sometimes extreme dynamic compression found on CDs, which is, by any definition, not compressed, bit-wise, at all?

 

[steve_1979]

I acknowledge that compression of dynamic range (reduction of difference between loud bits and quiet bits) is not the same as bit compression but I suggest that the one leads to the other.

Do you have any evidence to back up that suggestion?

 

[cheeseboy]

I acknowledge that compression of dynamic range (reduction of difference between loud bits and quiet bits) is not the same as bit compression but I suggest that the one leads to the other.

I may be out of date a bit but my understanding is that a digital number (1100101010101101 etc) is ascribed to each one of a sequence of analogue levels.

ok, not shooting you down in flames, but unfortunately the thought process doesn't work as dynamic range compression can be totally analogue and never even hit a digital circuit in it's life. (it's very rare that would happen nowadays, granted, but dynamic range compression was around before digital recording).

 

[Tannoyed]

Oh dear I seem to have stirred up a hornet's nest. I realise that dynamic range compression had been used in the analogue world long before the digital beast was awoken. Some of you may recall the dbx system many years ago which enabled both compression and expansion to be carried out. I have acknowledged that my understanding of this area may be limited and was seeking guidance on the matter. If dynamic range compression is carried out it would have to be perfomed before sampling the signal (or so I imagine) to limit the range of 'numbers' needed. Indeed CD's could equally well contain compressed information but what I hear routinely when listening to MP3 files is a compressed version of what I had previously heard on the cd in my collection.

MP3 is brilliant in what it achieves, and for use on the move it is great as far as I am concerned. In its best form it really is very good. Engineers have achieved so much for so long and with so little I expect that soon designers will be able to achieve anything with nothing!

I should be interested to know how they do it.

 

[cheeseboy]

Oh dear I seem to have stirred up a hornet's nest.

no at all

I realise that dynamic range compression had been used in the analogue world long before the digital beast was awoken.

yep, that was the point. Your train of thought was talking about digital, but given this is a pre-digital thing, then the rest of your train of thought was, well for want of a better expression, wrong.

If dynamic range compression is carried out it would have to be perfomed before sampling the signal (or so I imagine) to limit the range of 'numbers' needed.

nope, not really as it can be applied at any point really.

MP3 is brilliant in what it achieves, and for use on the move it is great as far as I am concerned. In its best form it really is very good. Engineers have achieved so much for so long and with so little I expect that soon designers will be able to achieve anything with nothing!

Couldn't agree more. Always amazes me what they can come up with

 

[andyjm]

Oh dear I seem to have stirred up a hornet's nest. I realise that dynamic range compression had been used in the analogue world long before the digital beast was awoken. Some of you may recall the dbx system many years ago which enabled both compression and expansion to be carried out. I have acknowledged that my understanding of this area may be limited and was seeking guidance on the matter. If dynamic range compression is carried out it would have to be perfomed before sampling the signal (or so I imagine) to limit the range of 'numbers' needed. Indeed CD's could equally well contain compressed information but what I hear routinely when listening to MP3 files is a compressed version of what I had previously heard on the cd in my collection.

MP3 is brilliant in what it achieves, and for use on the move it is great as far as I am concerned. In its best form it really is very good. Engineers have achieved so much for so long and with so little I expect that soon designers will be able to achieve anything with nothing!

I should be interested to know how they do it.

No need to apologise. You are quite correct in your posts.

At its most basic, compressing an audio signal reduces its dynamic range, which could then be represented with a fewer number of bits. So instead of using 16bits for 96dB of dynamic range, how about using 13 bits for 78dB of dynamic range?

16 x 2 x 44,100 = 1.41Mb/s

13 x 2 x 44,100 = 1.15Mb/s

A reduction in bit depth reduces dynamic range, but also reduces the bit rate of the audio stream. This is exactly what the BBC did in their early 70's digital distribution of FM stereo to transmitter sites.

So while compressing the audio signal (by reducing its dynamic range) can be looked at separately from compressing the audio data (by using data compression / reduction techniques), the two are linked.

 

[cheeseboy]

No need to apologise. You are quite correct in your posts.

At its most basic, compressing an audio signal reduces its dynamic range, which could then be represented with a fewer number of bits. So instead of using 16bits for 96dB of dynamic range, how about using 13 bits for 78dB of dynamic range?

16 x 2 x 44,100 = 1.41Mb/s

13 x 2 x 44,100 = 1.15Mb/s

A reduction in bit depth reduces dynamic range, but also reduces the bit rate of the audio stream. This is exactly what the BBC did in their early 70's digital distribution of FM stereo to transmitter sites.

So while compressing the audio signal (by reducing its dynamic range) can be looked at separately from compressing the audio data (by using data compression / reduction techniques), the two are linked.

I stand corrected. apologies tannoyed

 

[The_Lhc]

No need to apologise. You are quite correct in your posts.

At its most basic, compressing an audio signal reduces its dynamic range, which could then be represented with a fewer number of bits. So instead of using 16bits for 96dB of dynamic range, how about using 13 bits for 78dB of dynamic range?

16 x 2 x 44,100 = 1.41Mb/s

13 x 2 x 44,100 = 1.15Mb/s

A reduction in bit depth reduces dynamic range, but also reduces the bit rate of the audio stream. This is exactly what the BBC did in their early 70's digital distribution of FM stereo to transmitter sites.

So while compressing the audio signal (by reducing its dynamic range) can be looked at separately from compressing the audio data (by using data compression / reduction techniques), the two are linked.

I stand corrected. apologies tannoyed

No, you don't, just because a 16-bit CD has the POTENTIAL for 96dB of dynamic range it doesn't mean it's all being used (in fact it's probably NEVER all used), there are many CDs that have practically zero dynamic range (Metallica, Death Magnetic for example) but they are still 16-bit recordings. The dynamic range compression is not acheived by reducing the bit depth of the recording.

A lower bit depth recording would produce a lower *maximum* possible dynamic range but in practice it shouldn't make any difference to the vast majority of recordings. To use the example given previously there are few, if any, CDs that could boast a DR of 78dB. DRC in CD recordings is a deliberate engineering decision, it isn't an accident of bit-depth compression.

 

[andyjm]

MP3 is brilliant in what it achieves, and for use on the move it is great as far as I am concerned. In its best form it really is very good. Engineers have achieved so much for so long and with so little I expect that soon designers will be able to achieve anything with nothing!

I should be interested to know how they do it.

There are two styles of data compression. The first is data type agnostic (like a .zip file), which can contain any data. Basic compression is achieved by spotting patterns and repeated characters in the data and replacing these with counts. Rather than storing 50 zeros, it is much more efficient to store a code that says 'the next 50 characters are zero'. This is stripped out and replaced with the original characters when the file is expanded. These techniques are lossless (and much more sophisticated than my schoolboy example) and produce an exact replica of the file before it was compressed.

MP3 is the second type, and relies on understanding the content of the data (audio) and removing information that is redundant, reducing the data size as a result. Wikipedia has a good description, but the technique is based on psychoacoustic effects, where the ear/brain combination can only perceive a limited amount of information at any one time. Certain sounds mask other sounds, and if a sound is masked, there is no point storing it. By removing the masked sounds (and other techniques), significant reduction in data size can result. As the removed sound is not stored, it is lost forever, and as a result MP3 encoded files are lossy. It is impossible to recreate the original data file.

Edit: As for engineer's achievements, it reminds me of a quote from Colin Chapman (one of my heros) "An engineer can do for a shilling, what any damn fool can do for a pound"

 

[The_Lhc]

Whatever preconceptions people may have, the Foobar dynamic range meter doesn't lie and it says that WAV's converted to 320kbps MP3's have identical dynamic range to within 0.02dB.

Just out of curiosity what was a typical figure for the dynamic range you were seeing?

 

[steve_1979]

Just out of curiosity what was a typical figure for the dynamic range you were seeing?

It varies quite a lot depending on the tracks used (I tried it with classical music and the Red Hot Chili Peppers to test both DR extremes of music).

I've deleted the results from yesterdays tests but I've just run one more quick test to show an example. The track was picked totally at random (it's the first track from the Amadeus movie soundtrack) because this is the CD that was still left in my PC from yesterday.

Here are the results:

 

[steve_1979]

Whatever preconceptions people may have, the Foobar dynamic range meter doesn't lie and it says that WAV's converted to 320kbps MP3's have identical dynamic range to within 0.02dB.

This is a repeatable experiment that anyone who thinks that MP3 compression reduces dynamic range should try for themselves. I can even post a few screenshots of the results if anyone wants.

 

[steve_1979]

Oh dear I seem to have stirred up a hornet's nest.

Not at all. There's no reason why we can't all discuss our favourite hobby in a polite and friendly manner.

BTW this thread wasn't aimed at you. It's just something that I've noticed several times before on forums and has been bugging me for a while. It was purely a coincidence that it was your post that I picked up on.

 

[The_Lhc]

Just out of curiosity what was a typical figure for the dynamic range you were seeing?

It varies quite a lot depending on the tracks used (I tried it with classical music and the Red Hot Chili Peppers to test both DR extremes of music).

I've deleted the results from yesterdays tests but I've just run one more quick test to show an example. The track was picked totally at random (it's the first track from the Amadeus movie soundtrack) because this is the CD that was still left in my PC from yesterday.

Here are the results:

So that's what, about 12.5dB of DR?

 

[andyjm]

The example shown above actually shows the MP3 as having more DR than the original WAV version. With the majority of tracks it was usually the WAV that had slightly more DR than the MP3. But that's irrelevant because in all cases they were always matched to within 0.02dB which is far less that what's audiable.

Even with very low bit-rate MP3's (I tried as low as 45-85kbps VRB) the dynamic range was still always matched to within 0.06dB.

I thought foobar used the difference between peak and average RMS to measure 'dynamic range'. If that is the case, the numbers on your screen grab are in error.

 

[steve_1979]

I thought foobar used the difference between peak and average RMS to measure 'dynamic range'. If that is the case, the numbers on your screen grab are in error.

I'm not sure what process Foobar uses to calculate the dynamic range. It doesn't seem to say on their website as far as I can see. Looking at the figures and doing the maths I don't think it's the difference between the peak and RMS though.

Using the WAV left channel as an example. The difference between the peak of -7.00dB and the RMS average of -23.59dB isn't a 12.55dB difference.

(I'm not an expert so please excuse me if I'm making a silly noob mistake here )