Apart from switching format (Musepack .MPC for example), really MP3 is pretty much transparent (i.e. indistinguishable from CD when listening) with
lame --alt-preset standard -Z
(it's also pretty much transparent without the -Z, but that was recently introduced to fix a fairly rare but identifiable stereo-related artifact). Although it's about 192 kbps on average it's variable bitrate so can save bits where it's easy to encode, and use more when the music is tricky (so it's better quality than 192 ABR, which in turn is better than 192 CBR)
With Standard, lame's psychoacoustic model believes it will sound identical to the original file. Extreme, is there for people who feel a need to waste bits "just to be sure", and allows them to do so without worsening quality by using commandline tweaks, according to Dibrom who tuned the presets for 3.90.2. It copes with the placebo effect or ideas such as "my equipment is very good so it needs a little extra quality". In practice, Extreme makes very little difference to the rare artifacts left by Standard - and certainly not an improvement anywhere near commensurate with the extra bits it uses. Still, I'd be happy to listen to an Extreme file, because it usually sounds no different (i.e. no worse) than a standard file usually. Both standard and extreme often still sound pleasant even when they have artifacts, and you may not notice the artifacts if you don't know the music well.
If there are problems with Standard, they are usually limitations of the psymodel and of MP3 as a format. If you can't use Musepack (e.g you need to have hardware support) and have to stick to MP3, then Insane is the best that MP3 can do, and may manage to fix the problems in many cases.
An alternative is that many .MP3 decoders (MPEG-I layer 3) will also decode .MP2 (MPEG-I layer 2), which is not a transform codec but a sub-band codec. Decoders that decode MP2 include Lame --decode, Winamp's MPEG audio plugin, FB2K and reportedly quite a few (possibly most?) hardware MP3 players. MP2 has better transient time resolution (
as good as Musepack SV7) but isn't as bitrate efficient or as good at lower bitrates. However, it might do better on much of the music that trips up MP3 when it's used at 192 kbps or above (max=384 kbps), and might still use fewer bits than Lame Insane (=320kbps). MP2 psymodels aren't as finely tuned as Lame's, but with enough bits, they can reach very high quality.
@NIN9: As for filtering loud noises, it could be a number of things that cause the problems:
• You're using 192 kbps CBR. If you must stick pretty rigidly to 192 kbps for some reason, ABR is better quality, use --alt-preset 192 to achieve this, but better still, use Lame --alt-preset standard which is VBR and averages to something like 180-210 kbps over a range of most albums but uses more bits when it needs to and fewer when it doesn't to almost guarantee transparent quality)
• You're running into clipping on the decoder, in which case mp3gain (see my sig) will help avoid this without loss of quality by lowering the volume so it doesn't clip (and allowing you to equalize the volume of all files so they can be shuffle-played comfortably, for example).
• It might also be that you have an old, rather dodgy DAC that distorts badly near full scale, or you're running some amplifier to it's limits and causing clipping or distortion in the analogue domain.
• Also, some modern CDs are mastered near full scale as part of a loudness war, causing distortion, adding harmonics that take more bits to encode properly in MP3 and slightly muffling the sound.
• The other possibility is the placebo effect (which most of us have experienced). If you expect MP3 to have certain types of problems, you might perceive them to be there. If you can test blind - without knowing which is MP3 and which is CD - you might find you can't tell the difference that you were convinced you really heard before. Many of us have been surprised that problems we were sure we could hear vanished when we tried to identify them blind using ABX tools (like WinABX) or ABCHR tools.