Quote: Comparisons between image and audio compression are misleading and show a naive understanding of both approaches. Artifacts in JPG compression are caused by chroma subsampling and block-splitting, with the results processed by a discrete cosine transform and then quantized to allow for a more efficient lossless entropy coding. All of these approximations become obvious even in low compression ratios as you can test by simply zooming in to a JPEG which features curved gradients and high-contrast outlines. A JPEG is static, whereas an mp3 is not. The first step in mp3 compression involves breaking the uncompressed file down to samples and these samples to frequency bands which are then analyzed by a fast fourier transform. The FFT data is then run through a psychoacoustic model (think auditory masking) which further sorts the results into time slices, "windows" based on whether there's a steady noise or a transient. One so sorted the information is then fed through a modified discrete cosine transform which has the remarkable property of time-domain aliasing cancellation (i.e. removing artifacts because of a signal which changes over time, unlike that of a static image). The impetus behind lossy compression is how best to take advantage of the limitations of different human sensory apparatus. Your eye depends on detecting changes in chrominance and luminance but your auditory system depends on sensing rates of change of air pressure and minute differences in interaural-cross correlation. Whereas one can spend hours looking at a painting, photograph, jpeg or other static facsimile of visual stimuli with a magnifying glass and under different lighting conditions, you are not accorded the same luxury with the dynamic medium of audio. Actually it's even worse, since you are also subject to listening in an artificial and necessarily limited sound field where reflections and transducer limitations conspire to further distort the original signal (not that you'd ever perceive the "whole" signal because of auditory masking anyway). To address your final point, there has been quite a bit of research over the past forty years in developing a consistent vocabulary used to describe audio quality and even training programs for listeners to acquaint them with judging specific aspects of sound. The reason I'm calling you out in this spiel is because I want to stress that nothing is ever as simple as it seems and proper comparisons of signal processing methods require not only an appreciation and keen senses but also an understanding of the underlying physiology and mathematics.