The Distortion of Sound. The above documentary has seen over three million views on YouTube since its 2014 release. It was made in association with Harmon Kardon as a way to raise awareness of the evils of compression, citing MP3s and streaming music as the main culprits.
Unfortunately, like all great pieces of propaganda, it offers a few grains of truth that are then exploited as way to justify its highly dubious conclusion. Several of the statements made by numerous industry experts featured in this film are somewhat laughable.
Chris Ludwig, a chief engineer at Harman, describes MP3s as “dangerous”. Mixer/producer Andrew Scheps gives a description of how MP3s actually work that is clearly more psychotic than psycho-acoustic. Greg Timbers of JBL describes vinyl as audio nirvana, and then goes on to explain how MP3s remove all the sonic nuances from a recording.
The coupe de grace however is how the film demonstrates the deleterious effects of compression on digital music. If you don’t want to watch the whole thing, then start right here and watch the waveform go by.
Question: does the compressed version of Lianne La Havas’ sample recording sound worse because a) it has been shot through an MP3 algorithm or b) because of the overzealous application of dynamic range compression (DRC)?
We owe it to ourselves as audiophiles to talk about the word “compression” with a little more sophistication than the usual rhetoric of “MP3 bad, FLAC good”.
Let’s start with word “compression” itself. In the video, it’s short form for “data compression,” a field of information theory that has been around for decades and has single handily revolutionized the way we store and retrieve data.
Data compression’s first and foremost goal is to reduce the size of the original signal. Information theorists classify data compression algorithms as being lossy or lossless but do you know that the main difference between lossy and lossless is how data is classified?
In the lossless world, data is reduced by analyzing its statistical redundancy and then exploiting it to reduce its size. Over on the lossy side of things, data maybe classified as unnecessary in order to reconstruct the original signal, which incidentally is why lossy encoding schemes are not considered “bit-perfect” because some information loss can occur during the encoding process.
Although lossy encoding schemes throw away bits, it doesn’t necessarily follow that their data quality is really worse than their lossless counterparts. Much depends on the type and application of the data we are trying to encode.
For example, let’s say our target data is text, and we want to encode a document of plain text in order to reduce its size to save on storage. Would we use a lossless or lossy encoding scheme?
You might answer ‘lossless’ because if random letters get tossed from the original document (thanks Scheps!) the message becomes incomprehensible. But what if we apply an encoding scheme that analyzes a piece of text, measures the whitespace between each word and the paragraph boundaries, and reduces their size by a smidge? Let’s call it the “Lazy English Major” encoding scheme. Guess what? LEM is lossy. And also guess what? Depending on how aggressive we set LEM in reducing whitespace, we may not even notice the textual claustrophobia it creates. For the majority of message readers, LEM yields a 100% acceptable reproduction of the original signal despite being lossy.
Side note: I’d wager that if this LEM codec does indeed take off, audiophiles will spend countless hours online arguing over how LEM documents are far inferior to their lossless counterparts and that if anyone enjoys reading a LEM document, they are clearly missing out on all the hidden beauty that whitespace has to offer. Count on it.
Audiophiles might not find text data compression schemes all that exciting. How about audio? If we encode a piece of audio, would we rather it be lossy or lossless? As an upstanding audiophile, you holler “lossless”!
Not so fast. What if the audio we are encoding is cellphone speech? Would we really want our voice to be encoded to DSD before being transmitted over the air? If you do then you probably also prefer all of your voice mails to be pressed directly to vinyl once a month as well. Obviously, a lossy encoding scheme makes a lot more sense since the spectral content of speech is extremely limited and the medium in which it gets transported is generally unreliable. Once again, lossy FTW.
But what about music? Would we use a lossy or lossless encoding scheme to store and play back our favorite piece of music?
Any audiophile worth his/her salt will opt for a lossless encoding scheme like FLAC (or ALAC). But which sounds worse, Metallica’s Death Magnetic as a low bitrate MP3 or in FLAC? What about The Red Hot Chilli Pepper’s last release, I’m With You? Heck, feel free to compare the 24bit/96kHz high-res version to an equivalent Spotify stream.
I’ll help you out here a little: both albums have been subjected to significant dynamic range (DR) compression. Usually an artistic decision, DR compression is where the difference between the quitest and loudest sounds on a recording have been significantly reduced. The smaller the difference, the lower the DR score. Death Magnetic scores DR3 whilst I’m With You returns DR4.
For all of you folks out there who have patiently been waiting to remind me that you can indeed tell the difference between MP3s and FLAC (thank you very much), hold that thought just a little longer.
Assuming a quiet environment and thousands of dollars of audiophile gear, is lossy data compression audible with so much dynamic range compression applied to the recording itself. Feel free to start your race to the bottom in the comments section.
The sad truth is that no matter what format or streaming provider you choose, both of these records will come out sounding very poor indeed. The reason? High fidelity was not a chief concern during the recording process; volume was.
For these two albums, any gains in fidelity we might’ve otherwise heard from a strict adherence to a “bit-perfect” lossless encoding scheme are all but lost. Forever. No format, lossless or otherwise, is going to change that. So it begs the question: is dynamic range compression lossy or lossless?
Moreover, is the process of recording music lossy or lossless?
Almost all recordings see some sort of processing applied to them, whether that equates to heavy amounts of DRC or throwing a mix through a number of digital processors. Depending on what the engineer is trying to accomplish, not all bits will make it to the final cut.
Instead of having our data compression algorithm make the call on what bits are necessary and what bits aren’t, an engineer is now making those choices. Manually. The recording and mixing process, in the loosest sense, is lossy.
An album’s fidelity has a lot more to do with the decisions made by recording and mastering engineers than any psychoacoustic or linear predictive coding codec used to carry the digitally carry the recording from the studio to our hifi system or smartphone’s D/A converter. Don’t hate the player, hate the game!
Hopefully now, you have a slightly more refined view of the whole lossy versus lossless debate. Yes, I prefer FLAC. Yes, you should archive all your data in FLAC. And yes, most of the time, it doesn’t even matter that much since in the end, a lot of those bits were ruined with respect to high fidelity before you even had a crack at them. That’s why I believe we audiophiles spend way too much time infighting over different formats and delivery mechanisms, when our real focus should be on the true culprit behind the degradation of fidelity in digital music. And that is how those bits were made in the first place. Think about it.
You can read more of Alex’s thoughts on audio over at his own Metal-Fi.