Chasing digital music's 'codec killers'

If you've downloaded a Van Halen, Hole, Sarah McLachlan or even a Beatles song online, and it sounds just a bit off to you, you're not alone--you might have run into a "codec killer."

Online music services such as Apple Computer's iTunes and MSN Music are betting that digital music sales will rival or even surpass those of CDs one day. But to serve today's market, they're forced to use audio formats--called codecs--that transform recorded songs into relatively small computer files that can be downloaded over Net connections.

Most of the tricks audio engineers use to compress music into these small packages are almost wholly inaudible to the average human ear. But there are songs that prove consistently thorny for these technologists, throwing up audible defects whenever they are squeezed into Net-friendly sizes.

"You can fool most of the people most of the time, but you can't hide (flaws) all the time," said David Ranada, technical editor of Sound & Vision Magazine.

Although rare, the "codec killer" phenomenon is a potentially serious one for digital music backers. Just as the music industry had to assure consumers that the sound quality of CDs was as good as that of vinyl records, digital music companies must ultimately convince online music buyers that computer files provide a listening experience comparable to that of CDs.

Even just a few purchases of songs that wind up with defects, ranging from muddy sound to audible "artifacts"--sounds that aren't supposed to be in the recording--could turn off buyers, analysts say.

This idea of "codec killer" has been familiar to audio technologists since they started transforming audio into compressed formats. Much of a codec designers' job is figuring out how different technological tricks will interact with the ear and minimizing these ill effects.

Each compression technique uses different ways of packaging the data. But most modern song-squeezers essentially aim to remove from an audio wave information that won't be missed by the human ear.

Many of the tricks are based on using the ear's own physiology to fool it into hearing things that aren't really there, or figuring out which elements of a sound wave can be eliminated without substantially changing the sound. A succession of technologies such as MP3; Advanced Audio Coding, or AAC; and Windows Media Audio have gotten increasingly better at this during the past decade.

But these tricks run into problems with specific kinds of sounds.

For example, one common technique essentially divides a song into very small blocks of sound--a few dozen milliseconds, for example--and takes bits of information out of each block. The effect this has is similar to a blurry picture, smearing out the sound very slightly over time.

This effect is mostly indiscernible to all but the most trained ears. But some very sharp sounds--a wood block, or a castanet recording--often wind up with what's called a "pre-echo," or a hint of the sound before the full instrument actually begins.

That essentially confuses the ear, creating the perception of an unpleasant sound, said Microsoft audio architect Jim Johnston. He compared the ear's reaction to what the eye does when exposed suddenly to a bright light in the dark.

"Castanets are like turning on the sun at midnight," Johnston said.

Different issues create different audio artifacts, or flaws. Very pure sounds are sometimes hard to squeeze into a compressed state, in part because the same smear effect can undermine their purity. Songs that are recorded at very high volumes can prove difficult, as can songs with powerful low-frequency bass tracks.

"What's difficult for a particular codec depends on the codec itself," said Christopher "Monty" Montgomery, one of the lead engineers behind the Ogg Vorbis open-source audio technology project.

The worst effects could almost always be minimized by keeping a song in the highest quality format available, Sound & Vision' Ranada said. Changing from one compressed format into another--from iTunes' AAC into MP3, or from MP3 into Sony's ATRAC, for example--is almost always a bad idea, he said.

Even beyond the individual songs' difficulties, the prevalence of mediocre-quality MP3 songs on file-swapping networks and in personal music collections has had a damaging effect on music listening, Ranada added.

"I think people's tastes are being degraded by exposure to 128 kilobit per second MP3s," Ranada said. "If that's all you hear in your acoustical life, and never go to a concert, you'll never know what real music sounds like."