Comparison of digital compression codecs

[Dave2002]

I believe that amazon downloads are usually 320kbps mp3, while iTunes ones are 256 kbps aac - please feel free to correct me if I'm wrong. While I believe that generally aac outdoes mp3 for quality - that can only be if the compression bit rate is comparable. So, is there any objective evidence regarding the relative performance of these codecs at the bit rates given here?

[johnb]

The Amazon downloads that I've bought have definitely not been 320kbps - the average bitrate has been between 200 kbps and 260 kbps. (I believe they use a variable bitrate.)

[Phileas]

I think iTunes may use variable bit rates giving an mean of 256kbps (but I could be wrong).

[johnb]

I think iTunes may use variable bit rates giving an mean of 256kbps (but I could be wrong).

I believe Amazon also has a target average bitrate of 256 kbps (though mp3, not aac). However, in practice the average bit rate can be as low as 200 kbps, presumably it depends on the nature of the audio being encoded.

[DublinJimbo]

AAC files are generally higher quality and slightly smaller than MP3 files of the same song. The reasons for this are fairly technical (more about the specifications of the AAC format can be found at Wikipedia), but the overview of the reasoning is that AAC was created after MP3 and it offers a more efficient compression scheme, with less quality loss, than MP3s. Despite popular belief, AAC was not created by Apple and is not proprietary to Apple or its devices. AAC can be used with a wide variety of non-Apple devices.

— About.com

[OldTechie]

For iTunes have a look at http://images.apple.com/euro/itunes/...for_itunes.pdf . It has a very readable explanation of what it is all about. It is 256k VBR AAC.

[Bryn]

For iTunes have a look at http://images.apple.com/euro/itunes/...for_itunes.pdf . It has a very readable explanation of what it is all about. It is 256k VBR AAC.

Hmm. iTunes offers 256kbps VBR AAC (which particular version is unclear). Radio 3 HD Sound uses 320kbps VBR AAC-LC. So much for Apple quality, eh?

Oh, and I have a little Sansa Clip which is sort of aac compatible. Sort of? Well the only aac type files it will play are those created by iTunes. Take an AAC-LC or HE-AAC file and convert it to m4a with MP4Box and the resulting m4a will not be recognised by the Sansa Clip, though my JVC Car 'radio', which demands the m4a suffix for aac files, will play such files without a hitch.

[Dave2002]

Hmm. iTunes offers 256kbps VBR AAC (which particular version is unclear). Radio 3 HD Sound uses 320kbps VBR AAC-LC. So much for Apple quality, eh?

Oh, and I have a little Sansa Clip which is sort of aac compatible. Sort of? Well the only aac type files it will play are those created by iTunes. Take an AAC-LC or HE-AAC file and convert it to m4a with MP4Box and the resulting m4a will not be recognised by the Sansa Clip, though my JVC Car 'radio', which demands the m4a suffix for aac files, will play such files without a hitch.

There are several issues here, and it seems that we haven't uncovered all that is to be found here yet. My belief is that at like for like bit rates AAC will sound better than MP3. This is because AAC is (in some ways) a more complex codec, and incorporates several additional techniques - https://en.wikipedia.org/wiki/Advanced_Audio_Coding

Personally I have found it very hard to tell the difference between lossy compressed digital audio and uncompressed audio once the bit rate goes above 220 kbps. At lower bit rates it can be quite easy to tell the difference. I believe that above 256 kbps most people would find it very hard to distinguish. However, this will depend on the way the encoding is done, and I am assuming near optimum encoding. For preference I would tend to use lossless compression over any lossy method, though this is likely to at least double the file size compared with (say) 320 kbps MP3 or AAC. This could be mere superstition on my part, but at least should ensure that the data can be recoded if needed, and there should be no loss due to the original encoding. This is not possible with any source material encoded with a lossy codec. Repeated transcoding of audio which has been compressed using lossy codecs can certainly introduce artefacts which may become audible, even if a single encoding introduces no significant artefacts.

Putting all this aside though, I feel that there are additional factors for tracks acquired from commercial or other sources, or played back from streaming sites. Put simply, some of them are just not processed very well. Some audio enthusiasts thoroughly dislike any form of lossy compressed audio, and there may be several reasons for their dislike. One of them could be that they find that the quality is noticeably lower than it perhaps should be - and this could be because of a lack of care in the compression process.

Possibly Amazon's AutoRip facility may enable more considered judgements to be made, as users can compare lossy compression, either streamed or downloaded, with CD quality - from CDs, if they wish. I would expect in almost all cases the CDs to either sound almost identical to or better than any lossy compressed version of the same music. Things do get more complicated though if the CDs themselves contain audible distortion, which I have noted on some jazz CDs. Auditioning tracks before purchasing a CD can give an idea of what the CD will sound like, and the CDs, when they arrive, should sound better, but since there is inherent distortion in some recordings, the generally improved sound may still not remove some audio problems. There is probably no way of knowing this in advance of a purchase, except by reading reviews.

One possible exception to the "CDs will sound better rule" is in the case of tracks from iTunes which have been "Mastered for iTunes", where there is a claim that master recordings with more than 16 bits per audio sample are used for the encoding. Codecs such as AAC and MP3 are not constrained to 16 bits per sample, so if there is any benefit in having more bits per sample, this could cause the CDs to sound less good. Despite this, however, I and others generally still feel that any lossy compressed audio format will not achieve the same quality levels as CD.

Another exception might be for download sites where so-called high definition recordings are made available in various compressed audio formats. Linn have this feature on their site, and it is possible to buy CDs, in some cases SACDs, including multi-channel, FLAC and ALAC encoded Studio Master quality downloads, FLAC and ALAC encoded CD quality downloads (these will be 16 bit and lower sampling rate 44.1 kHz), and finally MP3 - typically 320 kbps compressed. Without knowing all the details, it is just possible that the MP3s could sound better than the CD quality material, because of the possibility that sample depth would have more benefits than any loss in the digital encoding. I have no experience of this personally, as I have only ever had CDs or SACDs from Linn. Despite the possibility that the MP3s from Linn could sound better, I'd still put my money on the CDs. Given a choice between CD or SACD from this or similar sources I will normally purchase the SACD version - unless there is a significant financial reason for sticking with the CD.

There are some sites where purchasing higher quality level downloads also allows lower quality, more heavily compressed versions, to be downloaded as well at no or little extra cost. Most of us are not going to purchase multiple versions at different quality levels if there are significant costs involved in order to test out theories regarding sound quality.

[Nevalti]

I believe that amazon downloads are usually 320kbps mp3, while iTunes ones are 256 kbps aac - please feel free to correct me if I'm wrong. While I believe that generally aac outdoes mp3 for quality - that can only be if the compression bit rate is comparable. So, is there any objective evidence regarding the relative performance of these codecs at the bit rates given here?

'Objective' evidence could presumably be identified by studying the wave patterns of a test tone but that still leaves the question of whether we can hear the different wave forms.

As you say in a later post, it is hard to hear the differences in sample rate as they increase, yet, very sensibly, you choose to buy CDs or SACDs rather than MP3s or AACs. Whilst it is indeed sometimes 'hard to hear the difference', compressed formats do usually reveal themselves many times in most movements and never in a nice way. A major factor in whether someone can hear the difference or not is what they are listening on. An old valve amp with worn out valves coupled with a pair of unrevealing speakers will be a very different experience from someone using electrostatic headphones straight off a DAC.

I believe I can hear the difference between CD and any lossy format I have heard BUT I have yet to hear any difference between CD & HiRes downloads. Is that my limitation or a general human limitation? My guess is that it is me. I WANT HiRes music to sound better but, to me, it does not - not on my current equipment anyway.

Off at a bit of a tangent, I usually find MP3 actively annoying rather like the LED tail lights which many cars have these days. I can see those damn LED tail lights flashing on and off ALL the time but I understand that many people can not see it at all; much the same as many people can't understand what is wrong with lossy codecs - not even the dreadful MP2 via DAB!

[Bryn]

... For preference I would tend to use lossless compression over any lossy method, though this is likely to at least double the file size compared with (say) 320 kbps MP3 or AAC. ...

Ah, but this is so very much content dependent. I have have, on occasion, found FLAC files that use less than 320kbps. Not often, certainly, but with some Feldman recordings, for instance, this is indeed the case.