i remember years ago, when mp3 was a newer thing, i had a big collection of mp3s that i would decode and then play in my car. very often it was obvious that these were originally mp3s. there would be a metallic sound to them and the bass would be off.

i know that advances have been made in encoding and i've heard accounts from many people recently that are suprised to discover that they can't distinguish between mp3s and originals on their ipods or portable systems.

last night i ripped tracks of several cds, encoded them with lame 3.9x at 192kbs, then decoded them back. i then burned a cd with these original and encoded/decoded tracks to compare.

i couldn't tell when playing them back to back, but my girlfriend was picking them out each time...that is, for the first 5 minutes of testing or so. after that, i think habituation might have set in because for the remainder of the evening she was unable to correctly pick them out as well.

my question is this then: i know that with very specific sound samples that are played over and over using ABX that differences can be spotted reliably, as long as you know what you're looking for and can compare fast enough.

but for general music, is there a bitrate which would be very nearly identical to the original, preserving stereo imaging, tonality, detail?

i am a bit of an 'audio enthusiast', so i am a victim of audiophile nervosa. when i know there is a difference, if it is very small and only noticeable when i listen very carefully, that knowledge eats away at me. i spent a good deal of money on my speakers and they are quite resolving.

You have to be careful because bitrate isn't quality, but filesize.

Problem samples don't just exist in "specific sound samples," but are found in general music as well. Somehow this makes me suspect that the likelihood of you noticing an artifact in general music would be low.

Stereo imaging, tonality, and detail are not very difficult to maintain with lossy compression. Whether/when MP3 would be adequate is a question only you yourself can answer.

You should stop thinking in bitrates. Some music parts simply need higher bitrates than others to reach transparency. You shouldn't aim for a specific bitrate, you should use one of the presets. Have you read the sticky thread with the recommended LAME settings?

For the vast majority of music and for the vast majority of listeners, --preset standard (or -V2 in newer LAME versions) will sound indistinguishable to the original.

so the music might not necessarily benefit from a higher bitrate? only portions of songs which may at a particular moment need that higher bitrate?

if stereo imaging, tonality and detail aren't affected by lossy compression...then what is?

The part that you can't hear. (With LAME aps, anyway . . . )

Yes. That's the whole reasoning behind setting a transparency standard for lossy encoders, because you may not be able to hear a difference at high bitrates. Quite a lot of music encodes down to the sub-160kbps range even with very high VBR settings, even stuff like symphonic music.

Here's my (admittedly not authoritative) answer.

All of these are affected, especially stereo imaging, but with an ideal lossy compressor, they are uniformly controlled by the desired quality level. ie, at a lower quality level all artifacts will become uniformly more noticable. The trick is to find the quality level at which most/all artifacts disappear to most/all listeners in most/all samples.

In most blind tests people succeed at here, the issues found really are just with specific sections, or even specific fractions of seconds, of songs. Things like air and detail and tonality are just not noticed as much, for better or worse. This is consistent with the goal of lossy encoders to handle the vast majority of music signals properly and transparently. Obviously there are some glaring effects, notably Vorbis's treble boost effect, which have been identified as common to most music pieces, but they are the exception rather than the rule.

Of course, some people can find such general differences at high bitrates. They're just harder to spot than specific trouble sections.