Saturday, December 5, 2015

On Losslessness

Recently, I've been thinking a lot about the nature of audio file formats and online distribution thereof. There seems to be a growing camp of people demanding lossless digital download options, but also a camp that claims that decent lossy compression is good enough for most people. Although I'm decidedly in the former camp, I would like to more thoroughly explore what the actual differences are between lossless and lossy compression.

Plenty of people have tried to determine if the differences are easy to hear. Generally, these analyses fall into two camps. The more populist surveys usually barely show a favorable outcome for the ability of an average listener to correctly identify a lossless file versus a lossy version. (See, for example, here.) The more specialized, audiophile studies fare somewhat better, although the specifics vary widely. Some people claim to be able to discern the difference with no difficulty, but these people tend to have high-end hardware and trained ears. (See, for example, here or here.) Most people can barely hear the difference, and it would seem even that requires more concentration and effort than is usually afforded during casual listening.

However easy it may or may not be to hear the difference, I am nonetheless interested in what exactly that difference is. The effectiveness of flac (the Free Lossless Audio Codec) in reducing file sizes to about half or two-thirds of uncompressed wav files should prove that some amount of lossless compression is possible simply by eliminating redundant data. Lossy compression also removes redundancies, but by definition also removes actual audio content to further reduce file size. The most obvious elimination is any frequency over 16 kHz, since many people cannot hear frequencies above that point, or cannot hear them well. Even I top out somewhere between 17 and 18 kHz.

After that, though, exactly what gets cut is not necessarily easy to describe. Fundamentally, information that is considered inessential is removed by the algorithm. However, some of this information may be detectable in its absence by careful inspection. To this end, I did some internet searching and found a few articles and discussions that address some common trends. Here are some of the conclusions I've come across:

1. Transients (e.g. snare hits) suffer. They get blurred, lose their sharpness, and may even acquire pre-echo. All forms of percussion can lose some of their natural punch. Such quick bursts of information are often too short for the codec's processing frame size and they get blurred across the frame.
2. Vocals lose focus and clarity. Our ears are particularly sensitive to the human voice and can detect seemingly subtle changes.
3. Cymbals and applause get distorted and rough. This is because high-entropy (i.e. "random" or rapidly changing) information changes too fast for the codec. This can sometimes also materialize as ringing or warbling.
4. Bass instruments get muddier. Lower frequencies have longer wavelengths, which can be longer than the codec's processing frame size, and thus do not get represented accurately.
5. Stereo separation and phase become distorted. Some of this is due to M/S (mid/sides) stereo mode, which instead of storing left and right, tries to reduce information redundancy by only storing the center (shared) and side (differences).
6. Dynamic loss and EQ loss is somewhat inevitable. Some sounds may get attenuated more than others, and the others may thus seem louder.
7. Noise (general murkiness, an underwater feeling, hiss, etc.) sometimes creeps in where previously there was desirable content.
8. Lossy compression can simply make things sound different, even if not necessarily worse. However, any deviation from the intentions of the artists and producers can reasonably be considered undesirable.
9. Lower-fidelity source material may actually suffer even worse, as whatever noise and other flaws exist in the uncompressed original may become exaggerated.
10. Genre, style, and the nature of the audio in question matter. Some types of music seem to compress better than others. Any reasonable audio comparison test should use a variety of types of music or audio.

There are, of course, a couple other factors to consider, such as the differences in acquiring and storing lossless and lossy audio. Hard drives are constantly getting cheaper and bigger, so the cost of storing lossless audio is a fairly marginal issue anymore. Acquiring the audio is another matter, although the difference there is also no longer as vast as it once was. New CDs are still only slightly more expensive than most mp3 stores, and used CDs are almost always cheaper. (The rip-and-resell approach has detractors but has been thus far legally unquestioned, at least in the USA.) Lossless online retailers are generally just about as expensive as mp3 stores, or at worst slightly more expensive than mp3s but still less than CDs. Hence, cost of acquisition is hardly a dealbreaker.

The real problem in acquisition is still that of availability. Lossless online retailers, while ever increasing in number and in content, still do not represent anything near all of the world's available music. It can be a pain to track this stuff down if it doesn't have the right type of following or industry support. There are a few significant websites (such as Bandcamp) and many individual indie labels (Sub Pop, Merge, etc.) and bands (speaking from experience: Ride, The Church, Wilco, and others) that offer lossless downloads, but many artists are still hard to track down.

This is also confused by the proliferation of HD retailers, which offer even higher sample rates and bitrates, despite that most people do not have the equipment to take advantage of the additional audio content. This wouldn't be a problem except that HD files are several factors larger and usually more expensive than any other digital format. (Only vinyl competes at that price range, and that's yet another story for another time.)

For me, lossless is the answer. While the quality advantage of lossless music may not be vast, the matters of file size and cost are less significant to me. The difficulty of acquiring lossless audio can still be a challenge, but it seems to be getting easier with time, and I am not opposed to CDs. In fact, if there is one issue that still gets me about most digital music downloads, it's the lack of album art. This is a big deal to me, and in fact was one of the first things I ever wrote about on this blog. Sometimes this can be found on discogs or other sites, but finding it in decent resolution is usually tough. If that hurdle can be crossed, then lossless digital downloads should clearly be considered the standard.


P.S. For the purposes of this discussion, I consider "lossless" to mean redbook audio CD quality, i.e. 16-bit, 44.1 kHz. HD audio is entirely other discussion with its own contentions, such as whether most listeners actually benefit from it, whether listeners can distinguish it, and whether the online retail options are any good. I do not have solid opinions of my own on these matters (yet).

No comments: