navigation background for audio comparison page navigation menu navigation menu email image riddle home to nigelcoldwell.co.uk



Bookmark and Share









Comparisson of Audio Compression

I've compiled a table containing the same audio sample compressed into different bit rate using several common compression techniques. Namely MP3 (constant bit rate and VBR,) windows media audio, ogg vorbis, AAC and flac. My intention is that you can use this to work out the lowest bit rate at which the audio sounds clear to you. this will depend on the equipment you're using, and your hearing. I firmly believe that if you can't hear the difference then there is no point wasting space.

I discuss re-encoding from one format to another below.

Notation:
k means kilo which is 1024. b means bit. B means Byte

Bandwidth
As you can imagine this page is quite bandwidth intensive, for me as well as you. I'd ask you to be sparing.
That is to say if you've decided that MP3 at 192kbps sounds the same to you as the original file then there is probably no need to download all the higher bitrates too. Finally i would ask that you try and only download the larger files once. Right click 'Save As...' and save them to your desktop. If you would like to download all the samples in one zip file there is a 57meg archive available on uploaded.net here (outside US only) or Google Drive here. Both are full speed downloads, you don't need an account or anything.

Details of the sample and the various encoding techniques used can be found below the table together with some thoughts.... Below the table is a Blind Test, it's been pointed out to me that some of the differences you can hear maybe because you already know, so it seemed worth while to create a test where you listen blind.

MP3 WMA AAC Ogg Vorbis FLAC WAV
32kbps 150kB CBR VBR 158kB 151kB
40kbps 186kB C V 187kB
48kbps 222kB C V 235kB 226kB 220kB
56kbps 258kB C V 260kB
64kbps 295kB C V 305kB 298kB 297kB
80kbps 368kB C V 379kB 375kB 370kB
96kbps 440kB C V 453kB 441kB 437kB
112kbps 513kB C V 510kB 509kB
128kbps 585kB C V 601kB 589kB 587kB
160kbps 731kB C V 749kB 734kB 740kB
192kbps 876kB C V 896kB 888kB 882kB
224kbps 1,021kB C V 1,038kB 1,013kB
256kbps 1,166kB C V 1,185kB 1,149kB
320kbps 1,457kB CBR 1,485kB 1,146kB
467kbps 2,110kB
968kbps 4,384kB
1.4Mbps 6,389kB


A Blind Test:
I've also added a blind test, the first 12 seconds of the sample has been taken from the 64, 128, 192 and 320 CBR MP3 files and converted back to wav so the files appear identical, the first 12 seconds of the uncompressed file is also included, each file is 2meg. Can you tell which one is which??? The full 37 second versions can be downloaded from uploaded.net HERE (outside US only) or from Google Drive HERE (30meg)

Blind Test File #1 File #2 File #3 File #4 File #5 Answer


The Sample (also WAV file):
The Sample contains three songs: Dido - My Lover's Gone, Daft Punk - Short Circuit and R.E.M. - Everybody Hurts the Tracks were ripped from cd using CDex 1.70 beta 2. And joined using Audacity 1.2.4.

The sample is 37.086176 seconds long.

MP3
The CBR MP3s were encoded using the Lame MP3 Encoder Version 1.32 engine 3.97 Beta 2 (MMX), constant bit rate (CBR), J-Stereo, quality factor 2 (high) this is my encoder of choice. The variable bitrate MP3s were encoded with lame.exe 3.97 with the default options left inplace apart from the --abr option. the target abr was adjusted to match the file sizes of the cbr.

Variable rate encoding adjusts the bitrate many times a second depending on the complexity of the sound currently being encoded. There are two methods of variable bitrate encoding available, one uses target quality and the other uses a target bitrate. The target quailty method will produce a consistent quality, the file size changing depending on the complexity of the sample, the target bitrate method will produce a consistent size, the quality varying with the nature of the sample. it is the target bitrate method used here, to provide a comparison to the CBR encoding.

The target quality method uses a notation -V0 (best) to -V9 (worst), i also encoded the sample using this method, the resulting file is directly comparable to a target bitrate file of the same size. The bitrates are given here: -V0= 232kbps, -V1= 200kbps, -V2= 182kbps, -V3= 152kbps, -V4= 138kbps, -V5= 126kbps, -V6= 114kbps, -V7= 98kbps, -V8= 89kbps and -V9= 66kbps

It's worth stating explicitly, if it is not obvious, that these quality level / bitrate comparisons are unique to this file, though the may server a a rule of thumb guide for other files.

WMA
Microsoft's Windows Media Audio encoder v9.2

AAC
This is not a codec i know much about but many of you will be familiar with it because it is iTunes' native format. the wav file was converted using 7.6.0.29, AAC, CBR. I haven't included apple's lossless format, simply because i do believe them if they say it's lossless. but the conversion resulted in a file of 4,081kB corresponding to around 882kbps. notably better than FLAC.

Ogg Vorbis
Ogg Vorbis 20051117

The audio was encoded using the quality factor technique rather than artificially specifying minimum maximum and nominal bit rates, as this is recommended for ogg vorbis. The quality factor (shown) was then adjusted so the file size most closely matched the mp3. it is then compared to the mp3's bitrate. The last Ogg file is a quality factor of 10, this is the maximum quality for an ogg file.

FLAC
The FLAC codec allows a compression level from 0 to 8. I encoded the sample with all of these levels which resulted in files with bit rates from 901kbps (level 8) to 968kbps (level 0). Since this represents a variation of less than 10% i have only included the one file of compression level 0. I should mention that the only difference is the encoding time.

Re-Encoding (Transcoding)
It is generally accepted that re-encoding from one lossy form of compression to another, (transcoding) is undesirable as each algorithm will have it's own set of rules when it comes to eliminating different aspects of the source audio, meaning you get the 'worst of both worlds' you have sounds eliminated by both codecs.

Worse than this, the second codec will waste effort trying to mirror the aspects of the first codec that may be undesirable. For instance if you have that familiar mp3 squelch in the first file the second codec must try to reproduce this in the same way that if you copy a video tape to dvd the dvd must use up valuable bandwidth in reproducing the VHS noise.

Whilst I've always taken this as fact i thought it would be interesting to demonstrate it so i converted the above 96kbps mp3 file to wma, then to ogg, then back to mp3. Compare the first and last files.

MP3 >  WMA > OGG > MP3

Conclusion
When not really a conclusion so much as some thoughts.

It's immediately clear that at very low bitrates the MP3 codec does not produce the clarity of WMA, AAC or Ogg Vorbis. This may lead you to conclude that the MP3 codec is inferior. However, the codec was not designed to work at this level, and you would never use it so. to form a valid comparison you need to simulate real world conditions and listen in your usual way....

One more thing it is also immediately apparent that the MP3 codec is infinitely better when used in variable bit rate mode. It is designed to work this way, and you do have to wonder why so many people encode in cbr.





© Nigel Coldwell