My last post was about the frequencies missing from MP3s.1
Today, I want to talk about bit rates.
To make sensible decisions about bit rates, it helps to understand a bit about how MP3 encoding works. When your computer makes a CD into an MP3, it has three main ways of making the music take up less space:
1) It throws out sounds you probably can’t hear—either because they’re “masked” by louder sounds, or because they’re only audible to a very small proportion of humans. Done right, this is an elegant exercise in efficiency. Done wrong (or too much) you music sounds tiny, thin and empty.
2) It describes the sound in terms of the shape of the wave, instead of as a big long list of values. If the wave isn’t a very complicated shape, it can do this with virtually no loss of quality.
Here are two MP3s of the same sound—a simple 440Hz sine wave. This is just about the easiest thing to make into an MP3. Although the first file is 10x the size of the second, they sound identical because you don’t even need 16kb to record 1 second of sine wave. Like a stick of rock, the file just says “440hz at -3dBFS” all the way through.
Here it is at 160kbps (mono).
Here it is at 16kbps (mono).
With our nice simple sine wave, there’s no extra data to throw out, so they sound the same. If we give it something really complicated, though, we’ll start to notice a difference.2
Here’s a bit of Debussy’s La Mer3, as a very high quality MP3.
That sounds pretty good to me, but here it is again, a tenth of the size.
Suddenly it sounds like it’s being played down a telephone. A file this size can happily hold a simple sine wave, but in trying to describe the complex harmony and sonority of Debussy’s orchestration, it has to make some cuts. All the notes are still there, but we’ve lost a lot of what is beautiful about it. When we’re looking for a bit rate that works for us, this is the outcome we’re looking to avoid.
3) Once the fat is trimmed off (1) and the important sounds reduced to their component waves (2), the computer looks for commonly-occuring patterns in what remains, so the information in them only needs to be recorded once.
In our first example above, that means saying “440Hz, -3bDFS” to define the single note, and “ditto” for the rest of the file.
With La Mer, the opportunities are less obvious, but if you see time in 44100ths of a second like the computer does, there’s plenty of repetition here. By itself, this third type of compression is lossless—you get exactly the same data out as you put in, but it takes up less space while being stored.
The combination of these three techniques allow us to make the files much, much smaller. Even the highest-quality MP3s are just a fifth of the size of the original files, but they can be much smaller.
The goal is to find the smallest file size that sounds good to you.
Let’s start with a 16kbps file. At this size, you could fit more than five days of music on a single CD:
I want you to make up your own mind, but I think you’ll agree that sounded pretty bad. This next one is twice the size, at 32kbps. This would let you put 54 hours of music on a CD.
This one is twice the size again: 64kbps. You’d get 27 hours of this on one CD.
Next is 128kbps, or eight times the size we started at. You’d get thirteen and a half hours of this on a CD.
Double that again, and you’re at sixteen times the size we started at. At 256kbps, you’d get six hours and 47 minutes of music on a CD. The Amazon MP3 store delivers music in this format.4
Finally, the highest bit rate supported by the MP3 format is 320kbps. That’s 20 times the size we started at, and 22% of the size of the original. You’d get about five hours and twenty minutes of this on a CD. If you buy music from ClassicsOnline this is what you’ll get.5
You can try all this with your own music, indeed I’d encourage you to. Hook up your computer to your stereo, make some MP3s (and other files), shuffle them up, and try to tell them apart. Remember: bigger is not always better. If you want a fast car, you don’t buy the one that uses the most petrol. You buy the one that goes fastest. If you’re looking for an audio format that sounds good, don’t go for the one that uses the most data. Go for the one that sounds best, and have fun.
I originally (and wrongly) wrote that the Amazon MP3 store delivered 320kbps files, not 256kbps files. This has now been corrected.
A typo in one of the footnotes said of Variable Bit Rate encoding “there’s no good reason to now use it” which is the exact opposite of what I meant, which is “there’s no good reason to not use it.”
1 All the files on this page are MP3s. This has become the format of choice for many mainstream download stores (including ClassicsOnline and Amazon) because it works on almost everything. Many of the same basic principles apply to both AAC (used by iTunes) and Ogg Vorbis (used by Spotify). These are both more sophisticated formats that avoid some of the more complex inherent weaknesses of MP3 at the expense of ubiquitous compatibility with all players. In general, either AAC or Ogg Vorbis should sound better than MP3 at a given bit rate, so if getting the best possible sound out of the smallest possible file is a priority for you, I’d suggest you check them out.
2 Simple sounds are easier to encode than complicated music, so it tends to be that you only notice that you’re listening to encoded music when something complicated or sudden happens. It’s in these places where the bit rate isn’t high enough. To overcome this, modern MP3 encoders use “variable bit rate” encoding, where a small amount of data is used for the easy bits, and a lot of data is used for the difficult bits. It averages out at the overall target bit rate. I haven’t addressed variable bit rate encoding in the main body of this post because it’s pretty ubiquitous now, and there’s no good reason to not use it.
4 iTunes also delivers 256kbps files, but in the AAC format1.
5 If that doesn’t sound good to you, then you might like to try theclassicalshop.net, eclassical.com or hdtracks.com—all of which will sell you full CD-quality downloads of just about any Naxos record, and many of those from the labels we distribute.