CS70 wrote:The difficulty with these things is that it's easy to overgeneralize - the devil is often in the the details (i.e. the context of usage).
Absolutely. Complaining about (or praising) MP3 is meaningless without also specifying the bit-rate (and to a lesser extent, the coding scheme) being used. The different coder designs aren't all equal either (see below), so you really should identify the specific coder being used, too.
High bit-rate MP3s are very hard to distinguish from the linear PCM source, by design. The proportion of people that can reliably detect artefacts is intentionally -- and demonstrably -- very, very small. The peer-reviewed research has been done and confirmed.
Conversely, low bit-rate MP3s tend to have more obvious artefacts that many will hear quite easily -- although a lot depends on the nature of the source material and how well it can mask the coding artefacts. A well-recorded talking book (single male speech) recording, for example, encoded at 128kbps may well sound perfectly acceptable and artefact-free, whereas a dynamic orchestral concerto at the same rate will probably sound pretty poor to many people.
Sure, the science is based on certain assumptions but these assumptions (and the corresponding algorithms) have been tested, verified and adapted several times over years before the format was finalized.
The really clever aspect to the MPEG codecs is that the specification only defines the way the replay decoder works. That intentionally leaves the door open to continual development of the encoders, so that as the understanding of the human auditory model improves, masking decisions determining which parts of the audio content are 'irrelevant information' and are therefore discarded, can be further optimised. All the encoder has to do is comply with the output data formatting expected by the receiving decoder. So different coder designers can make different decisions and prioritise different aspects in their algorithms.
And that's why certain people can hear stuff and other don't.
It's the familiar Gaussian or normal distribution model. The vast majority of people's hearing acuity sits in the middle of the curve. As you move out towards those with better or poorer hearing than 'average' their number rapidly diminishes, but there will always be some who are significantly better or worse than the 'average'. I don't know the exact percentiles for the different MP3 bit-rates but, it is designed to work for most people, most of the time... and it does that very well.
After many years of adoption, it seems that with a good encoder, a 256/320kps bitrate is definitely good enough, and 128kps holds water in many cases for most people, as theorized. No surprise most people don't complain!
I think that's a very good generalised summary. :-)