I’ve been meditating about audio fidelity recently, under the influence of The Civil Wars and Jeff Atwood.
What happened was, I kept catching videos and radio spots by The Civil Wars and liking them. I thought maybe I should buy some, so I visited their Web site and noticed with pleasure that they sell uncompressed FLAC, not just MP3, so I snapped up Barton Hollow. It’s good. But am I fooling myself in spurning compressed music?
Harmony · The Civil Wars are all about that special thing: girl voice and boy voice sing together. Human vocals don’t ask that much from an audio system; the notes don’t go that terribly high or terribly low, nor is the dynamic range huge. Production techniques are simple: use a good microphone, put the vocals somewhere near the center of the soundstage and mix the instruments in behind them. What you get depends on the song and the singing; which is as it should be.
So it’s not obvious that The Civil Wars’ music benefits from my insistence in paying the extra disk space FLAC costs. Some would go further, and argue that no music does.
Bitrate Horror · Coding Horror is the handle of Jeff Atwood, whose accomplishments include stackoverflow.com and a lot of really good blogging. He thinks 160kbps MP3 is plenty good enough, and attempted to prove it in The Great MP3 Bitrate Experiment, which is worth reading if you haven’t already.
He actually provided some audio samples at a variety of bitrates and invited people to pick the best. Most couldn’t. So he must be right, right?
I went and got them, and listened carefully on my audio system, which is moderately high-end. The bits come out of a Mac via USB into the highly-regarded Benchmark Audio DAC, into an old but very good Linn Kairn preamp, another old but very good Simaudio power amp, into not-that-old and really excellent Totem speakers and subwoofer, playing (and this really matters) in a large rectangular room with a hardwood floors and enough doors and bookcases to control standing waves; I’ve measured the response and it’s really not bad.
I couldn’t tell Jeff’s samples apart, except for I could tell the “worst” one (128kbps) was subtly different; not sure whether better or worse. So, Jeff must really be right, right?
I actually don’t know. I’m prepared to believe, especially given the advances in encoders tuned to human hearing capabilities, that there’s a huge fall-off in perceptible quality at some point on the bitrate curve, and that it might be somewhere around 160kbps. But Jeff’s samples are not an effective tool for investigating this issue. They are the opening minute or so of Starship’s We Built This City. It’s bad music that no human with good taste would actually listen to for pleasure, on top of which, as far as audio quality goes, it’s a turd.
There’s no deep bass and no illusion whatsoever that real musicians are in a real space. The voices are heavily processed and the instruments are a big cheesy splodge of 80s synth noise. Now, as for the drums; listen to when they come in, there’s this big pan across the sound-stage, right to left. Do you think this is the drummer working his way around his kit, with painstakingly-set-up mikes to catch the sound moving in space? Me neither. It’s either some sort of primitive drum machine or a mono capture, and the guy at the mixing board is twisting the pan knob. That mixing board, given the date, is probably an eighty-channel monster, each channel having like ten different cheapo controls. Not exactly audiophile values.
After a few painful minutes with Starship, I switched to The Civil Wars and Barton Hollow. The difference was epochal; I was hearing acoustic instruments reproduced with a lot of soul, actual human voices including breath and chest subtleties, and solid location of the voices and instruments in horizontal space. Oh, and songs that touched me.
I’d call Barton Hollow competently but not brilliantly produced. The voices are mixed a little too far behind and among the instruments; they deserve better highlighting. And any echo from the walls of the studio is brutally suppressed, the voices and instruments are surrounded by deep black digital silence. It sounds good, but not like being in a real room with real musicians; no you-are-there there.
Anyhow, if Jeff had extracted samples from something produced at least as well as this, and I still hadn’t been able to spot any difference, I’d be starting to line up with him on bitrates. But given that the music was recorded at much lower resolution than that of a decent home-audio system, I can’t believe any useful measurement occurred.
Things I believe · About audio reproduction:
Music quality trumps audio quality; some of my favorite records sound like dogshit.
For example, on Barton Hollow the best song by a mile is the title track, even though the guitars and some vocal parts sound kinda compressed. Because the words and music clutch at your heart. Well, mine anyhow
The most important link in the audio-quality chain is in the studio; the initial recording and production phase. If you produce a stinker like We Built This City, nothing down the chain is gonna save it.
There is a lot of snake oil and bullshit in audiophile circles. Many of the claims are in the ridiculous-to-physically-impossible range.
A solid high-end system, carefully set up in a decent room, sounds dramatically, stupendously better than ordinary big-box packages, on both well- and poorly-recorded music.
It’s really hard to isolate why high-end sounds better, because the claims of its proponents are often bullshit, and this has irritated the skeptics so much that they tend to dismiss the whole pursuit.
It’s a good idea to buy and keep uncompressed music, because when we actually get a good engineering handle on whatever it is that makes high-end systems sound good, we can maybe optimize for that and get more out of the signal, unless you’ve thrown too much of it away by compressing it.
I have a medium-large music collection; eleven and a half thousand songs, 33.5 days of music, and it’s almost all uncompressed. It occupies less than 250G of disk space, the cost of which is negligible. Why would you throw away musical data that might prove to be useful, to save pennies?
Musicians who want me to open my wallet should operate a competent Web site with a store that offers uncompressed downloads.
That Boy-Girl Thing · If you like it as much as I do, you might want to check out any of the records of elderly Greek folkie Georges Moustaki. Also, Emmylou Harris and Gram Parsons on Sleepless Nights, Emmylou Harris and Bob Dylan on Another Cup of Coffee, Emmylou Harris and Neil Young on Wrecking Ball... spot a pattern?
Happily, the form is not dead, as witness The Civil Wars. Also, catch this 2009 YouTube of Feist and Wilco. Feist and Tweedy are remarkable; their unison is so exact that it sounds not like harmony, but one voice with two parts. There doesn’t seem to be a well-produced uncompressed version online... but anyhow it’s the music that matters.
Comment feed for ongoing:
From: Rob (Sep 22 2012, at 03:32)
You forgot "All the Roadrunning" by Mark Knopfler and whatshername.
Incidentally, I caught her live one time, she did "Michelangelo" (which I suppose refers to Gram Parsons), and it made me cry. She is remarkable.
From: Salvatore (Sep 22 2012, at 06:27)
Do you have any thoughts on purchasing and storing "studio master" or "high definition" music files?
From: Eric Knapp (Sep 22 2012, at 07:47)
I heartily agree that high quality recordings will sound better in flac. It was this recording that convinced me: http://www.andrewyork.net/Discography/YamourFlac.php. Highly recommended if you like modern classical guitar.
I also agree about music being more important than the recording quality. Back when I was in my audiophile stage, it was frustrating when musician friends would listen to our fine systems. They couldn't hear the sound system, they only heard the music.
From: Nick Carr (Sep 22 2012, at 08:13)
"We Built this City"? Jeez. The best compression technique for that would involve the use of a steamroller.
But, anyway, I see level compression as far more insidious than data compression when it comes to the listening experience.
From: J Chris Anderson (Sep 22 2012, at 08:18)
I encourage you to check out SACD. They do digital audio differently, instead of recording the volume of each sample, they record the delta between now and the previous sample. And they do it at an extremely high sampling rate so that the overall data rate is similar hi bit rate PCM audio.
So the response curves of the medium are somewhat similar to vinyl, it has less available dynamic range for high frequencies. But the highest encodable frequency is much higher than the nyquist cut offs from CD audio.
The most important aspect for somebody with a listening set up like yours, is that the ultrahigh sampling rate means that it can capture differences in arrival times for stereo-recorded audio that you just can't capture with PCM encoding.
Of course this only works if the studio and production chain responsible for what you're listening to is either high fidelity analog, or else also uses SACD DSD style encoding.
From: Ant Bryan (Sep 22 2012, at 12:36)
I agree with most of what you say...have a high end system, 3 TB of losslessly compressed music, and might not be able to tell the difference... ;)
except that the terms are off.
you want losslessly compressed music (flac) if you have the choice, or higher quality lossy compression files (mp3). nobody wants uncompressed music (wav or aiff).
From: Dave Walker (Sep 22 2012, at 13:03)
Nice setup :-). I still swear by (rather than at) little silvery discs, but invoke more considerable tweakery in the path between transport and DAC. My little setup's mostly old Meridian 500 series kit, with a pair of 557s driving into very weird, hellish to drive, but rather gorgeous Dali Skyline 2000s.
I haven't looked it up for a while, but the most interesting audio format I've heard is DVD Audio (DVD-A) encoded with Meridian Lossless Packing (MLP). It's designed for music in a 5.1 format, but there's always a high-quality stereo mix down on the disc too.
I haven't heard a great deal of boy-girl vocal (will have to check your favourites out), but in terms of really good girl-only vocal with excellent instrumental performance and recording quality, I recommend "Waiting" (from "Mo-di", by Mouth Music) and "Way Down Deep" (from "The Hunter", by Jennifer Wearnes). They'll give your bass drivers a gentle workout, especially the Celtic drum on the first one.
From: JulesLt (Sep 22 2012, at 16:32)
I can think if some specific recordings that have sounded appalling in MP3 but I am unsure if that is down to bitrate or poor encoding software.
Cymbals in jazz seemed particularly problematic.
Equally, if I compare an original CD with an AAC played back through the same amp (via an Airport Express) I can clearly hear a difference.
Which again may be more down to the CD player having way better DAC and amplification circuitry than a box that also wants to be a WiFi extender and print server.
It may of course be true that most people can't tell the difference. Certainly not enough to care about it.
But I would be intrigued to see the same test performed with different encoders / bit rates / codecs, on something challenging.
(And I think you have something that it needs to be a recording where instruments have a clear 'space').
As for boy-girl harmonies, my big one over the last year had been Two Wings - Feet.
There's some lovely earlier stuff on YouTube when it was just Ben Reynolds and Hanna Tuuliki, rather than a group, which have not had a proper recording - 'Love's Memory' in particular.
(On her own, Hanna's voice can be a bit much for some, but it harmonises well with Ben's more conventional approach)
From: Bradley (Sep 22 2012, at 16:47)
Some comments on bitrate, in general and not specifically targeted at anyone:
I would posit that aside from there being a myriad of other factors contributing to quality for a given piece of audio, bitrate is not a reasonable estimation of that quality. It really only addresses quantity of data allotted over time. More efficent codecs will pack in more quality per bit and even then, will each have their own architectural flaws that break down in certain scenarios.
In the video world this is much more apparent and projects like x264 eschew the use of bitrate as a general quality estimator, instead relying on "rate factor" for non-broadcast purposes, and even that has its limitations (e.g., animation versus film content).
All that to say: bitrate for a given, specific codec may be a reasonable estimation of quality if the original material's type and composition is taken into account, as well technical factors like recorded bit depth and frequency. This is rarely the case. So when people talk bitrate, I often think, "You're doing it wrong."
My (unsolicited, ha!) advice is for everyone to choose a codec that is compatible with their playback devices and find a bitrate or quality setting that they find to be mostly transparent for what they listen to. Maybe that's 256kbps AAC because they're Apple people. Maybe it's 160kbps MP3 because it sounds fine and plays in their car stereo. There are a lot of variables and no silver bullets. I for one, really do not care what everyone else uses. With lossy codecs there is no best, only what one finds acceptable. And that may change over time as people listen to new music and hear how different audio systems reproduce it.
I consider myself an audiophile and prefer lossless audio for music. I'm more than fine with compressed audio for video. Naturally, I'm me and everyone else is everyone else. Whatever you enjoy, enjoy it.
Aside, my understanding is that DSD has already made its way into certain DACs, (AKM and XMOS come to mind), so we're mostly waiting on audio subsystem support in computer operating systems. A DSD over USB standard is gaining support, which you may read about here: http://www.audiostream.com/content/usb-link-dsd-audio-pcm-frames-andres-koch
From: Paul Hoffman (Sep 22 2012, at 19:36)
So, why are you not reproducing Atwood's experiment. Wouldn't producing those MP3s take less than an hour? I offer to take the properly-labelled MP3s and give them blinded names, and retain the linkages. Maybe do this for some male-female vocals, some well-produced classical, and so on.
From: Meower68 (Sep 22 2012, at 21:31)
I have nowhere the equipment you do and, with my smartphone and a decent set of earbuds, I can tell the difference between 128 and 196 kb/s MP3. As you suggested, though, I use different songs.
"The Launch," Boston, Third Stage. The sheer dynamics of this is murder on any codec. Of course, I have 2 copies of this album on vinyl, so my ears are very familiar with it.
"Hotel California," Eagles, Hell Freezes Over. I got my hands on a DVD of the opening concert from this and was stunned by the pure clarity of this, particular version of the song. The CD is almost that good. I've actually been tempted to acquire some SACD hardware, particularly for this song.
I encode my MP3s at 192 kb/s because I can't tell the difference between that and higher levels with any equipment I own.
From: Kevin Marks (Sep 23 2012, at 00:05)
One argument for actually uncompressed audio, as opposed to losslessly compressed Flac, is that it is far more resilient to bit errors. Yes, it take up twice as much space, but flip a bit and the chances of it being audible are low. Flip a bit in Flac and you lose a whole audio frame.
Famously, mp3 was tested on Suzanne Vega singing Tom's Diner (a capella). If they'd tried harmonies, they may have done better.
Also, you should look up Gregson and Collister
From: JulesLt (Sep 23 2012, at 02:18)
I wonder how prevalent bit errors are in the real world - if you believed the cable nerds, getting a picture to your TV over a £5 HDMI lead should be near impossible.
That would also be an interesting experiment - at what level do individual bit errors become distinguishable to the ear? And what is the actual error level.
Your high end audio CD players talk a lot about error correction - are they saying the physical side of the player is more error prone than the CD player in a Dell laptop, given that a single wrong bit in a zipped software install can be very bad news?
Hell - we worked out a way to accurately transmit bits down the same wires as analogue voice calls.
So my guess is anything on the digital side of the DAC is generally reliable at accurate storage/transmission of bits.
From: example (Sep 23 2012, at 05:03)
Don't forget that people lose their ability to hear higher pitched sounds as they get older. And the thing is, higher sampling rates help preserve higher frequency data.
I would imagine that higher frequency stuff is the first 'casualty' of reducing MP3 bitrates (of course they don't actually work the same as sampling rates). Those changes might not be noticeable to older people who have lost the ability to detect high pitch sounds.
From: hawkse (Sep 24 2012, at 04:34)
Thanks for the tip; Barton Hollow is great!
From: doug k (Sep 24 2012, at 10:40)
I'm happy to hear you too found Jeff's samples mostly indistinguishable.. I can't afford audiophilia but thought I would be able to discern the difference.
Xentrax posted the gizmodo bitrate test music in the first Coding Horror thread,
These samples are what Paul Hoffman is looking for, and are much better for all purposes than cheesy 80s shlock..
From: len (Sep 24 2012, at 13:48)
As to expertise, see Neil Young and T-Bone Burnett about bit rates and fidelity. Young discussed it in a recent NYT interview. Burnett is on record on the topic in several places. It is a good idea to understand what the music producers are saying because ears vary by training.
It is also interesting to read an article on the listening quality of music as a function of bit rate that can be afforded by the consumer that doesn’t account for the costs of high fidelity recording in a market where the technologies that enable piracy consume the resources required for high fidelity recording.
It illuminates a philosophy of philosophers who put sticks in their eyes so they can see the truth more clearly.
You get what the musicians can afford to make. Everything past that is a technology hustle.