What Data Compression Does to Music (2012)
soundonsound.com
soundonsound.com
This is a somewhat older article, but most lossy encoders at the time were still very good even at fairly low (sub-96 kbps) bit rates.
I just really don't understand this widespread need to believe "lossy" doesn't really mean "lossy." It's a psychoacoustic model and people are different, and further you can (as demonstrated by the article) even show where various encoding artifacts are feasibly audible to plenty of people. I just don't get it.
I often think I can hear the difference between lossless and lossy, but IIRC, the last time I did ABX testing on that, there was only 1 out of 5 tracks I tried where I could tell the difference at a statistically significant rate. (And I'm not sure I'd be able to repeat that.) The best argument I have for storing music in FLAC (or ALAC or some other open lossless format) is analogous to the argument for storing images in a lossless rather than lossy format if you have the space: if you ever have to transcode the file to something else, start with the highest quality possible.
Hate to sound like a broken record, but what are you arguing with me about?
I certainly agree that artifacts in 128kbps MP3s are noticeable but the assertion that 320kbps AACs or MP3s created with modern encoders are audibly different from uncompressed audio really needs some evidence. Every single double blind test I have seen conducted between them has shown that nobody can tell.
I can most certainly tell the difference between 320kbps lossy and red book audio on well done recordings.
Ex at 60db detecting 100hz sounds is easy, at 20db it’s outside of human hearing range.
After all of that, though I had double blind evidence I could hear a difference, I knew I didn't really care about that difference. I couldn't even describe the difference - it was almost subliminal. And that was with a harpsichord, which is a known difficult sound to encode. I've been a lot more comfortable knowing lossy encoding isn't destroying my music enjoyment since those tests.
The interesting questions to me:
- Can a difference be heard?
- If yes, are people able to reliably identify A vs B (as in, can people identify whether it's A or B rather that whatever they listen now is not the same as the last they heard)
- If they can identify a difference, which sounds better to them?
- Do different headphones give different results?
- Same question with headphones vs speakers?
There are absolutely huge differences between budget consumer and high-end professional equipment. There are also huge differences between high-end audiophile and professional equipment - the former being designed to flatter and enhance the sound, the latter being as neutral as possible.
A good few years ago a high-end audiophile CD/DAC maker was caught deliberately rolling off the top end on their equipment to make it sound smoother and "less digital."
It wasn't accurate, but a lot of people liked the sound enough to pay five figures for the boxes they made.
Professional audio has a different goal. The aim is to create a mix that "translates" - sounds good on as many different systems as possible, from budget earbuds to audiophile.
For a long time studios used a pair of shitty Yamaha bookshelf speakers called NS10s next to their ultra-expensive reference speakers. NS10s were almost the worst speaker ever. But if a mix sounded good on the NS10s, it sounded good on anything.
The question of preference, though, is what matters, and that's where I am now firmly convinced lossy compression is fine for me: the sort of differences that I could hear made zero difference to my enjoyment of actual music.
128 is usually pretty easy to pick out, once you know what to listen for. There's no soundstage. They just sound flat.
The tricky part is distinguishing between 320 and uncompressed. 320 is pretty darn good. Even when I can hear the difference, if it's not a song I'm familiar with, I can't tell which one is better. It's just different.
The one I got completely wrong was the Jay Z track. Even re-listening knowing which is which, I still can't tell the difference :( - but this is a style of music I rarely listen to, and maybe that's the reason.
It's also a fair few years since I was 30, and that could be an alternative explanation.
(I'm actually quite surprised I did as well as I did, as the 128Kbps versions weren't obviously worse, and it did take a few listens on headphones. I should probably do another blind test with a 2003 build of LAME or whatever, not so much to compare MP3 vs WAV, but more to measure how much my hearing has deteriorated over the years...)
> I barely ever can. > It's not something I'll notice when listening for enjoyment.
Pretty much tell the story.
The explanations aren’t themselves incorrect — there’s certainly low-pass filtering, pre- and post-echo artifacting and so on, and these reflect in the waveforms — but to reference a waveform and to say “this sounds bad because it looks like this” doesn’t really hold water. Visual differences are not necessarily audible differences, which is the fact psychoacoustic encoding leverages.
If an encoder’s design goal was to produce visually-comparable waveforms, a different approach would be used.
> In fact, CDs can only replicate frequencies below 22kHz and even that's within the hearable range of many (mostly younger) people.
There is no useful or meaningful content above ~18 kHz (aside from dither) on most recordings. There may be exceptions, but by and large, on real music recordings, the information above that frequency isn’t valuable, and is the among the easiest things to discard.
I can’t imagine if you compare a 192kbps MP3 with a DSD256 audio file played on reference monitors from a high performance DAC there’d be no audible difference. Personally, using a McIntosh integrated amp and Focal Utopia headphones the first time I listened to a DSD track, 1-bit word depth sampled 2.8M times per second, it was unlike anything I’d ever heard before. I acknowledge that’s not a direct comparison, but all i’m saying is maybe the lossy vs lossless test performed on a low fidelity audio chain where the signal is always meaningfully subject to harmonic distortion before you hear it is the reason most people can’t tell a difference?
The idea you can take a group of non-professionals and run tests with no requirements on equipment quality and then draw conclusions about what any human can hear is absurd.
Those tests (Hydrogenaudio was one of the main proponents back when lossy codecs were more important) are a decent way to tell if something is audible for the average listener with average equipment. It cracks me up when people start telling me what I can or cannot hear based on those tests without knowing anything about me or what kind of equipment I own. This happened just the other day on an audio forum.
Another dimension to this is that DAC quality has been increasing steadily while the prices of high fidelity DACs have been dropping. The DAC chips on the market today are really the best ever made (ESS and AKM notably). More people than ever have access to (near-)reference quality DACs. When lossy compression tests were popular few people had access to reference quality DACs. I remember people talking about using their computer sound cards as sources for those tests.
I think a lot of people react negatively to this topic because it's considered elitist (expensive toys). The good news is just about anyone can afford a near-reference quality, inexpensive headphone setup these days. You have to do your homework and read reviews but they definitely exist. I recently picked up a DAP and IEMs for ~$300 total (for running) and it's 90% of the quality of my reference rig. I listen to it instead of my reference rig sometimes. It's that good, despite the price.
(Even on that, I can easily ABX FLAC and 192kps AAC, which has been universally declared "transparent" more than once.)
If you think your smartphone's audio jack sounds good, you really need to listen to a device with a proper amount of output power (you need more than you think for headphones) and a high resolution DAC. You need both to be good, though they can be part of the same device. When you have them, music really becomes holographic and, for lack of a better word, alive.
Just disprove them with data! Until then, casually mentioning "award-winning" acquaintances and telling people they need "proper" gear reads a bit condescending and probably turns some people defensive.
I only take issue with people telling me what I do or do not experience -- because of the absurdity of that. The fact that this is about audio is coincidental.
I completely missed the Hydrogenaudio multiformat test in 2014. I wonder how much things progressed since when I participated in the early tests in 2006.
Headphone amp: http://nwavguy.blogspot.com/2011/08/o2-details.html#resource...
Which ones did you choose?
The IEMs I struggle to recommend: Fiio FH1, which is a dynamic driver / BA 'hybrid'. They are enjoyable to listen to but definitely not reference -- bass heavy. They're leagues beyond a typical earbud, however, and for $75 it's hard to complain.
I mainly don't recommend them because they're the first IEMs I've tried anywhere near that price range, apart from freebies included with smartphones, etc. They might be easily bettered by something else in the same price range, especially since there's a lot of competition there now.
I was intrigued by this paragraph in the article:
>> Psychoacoustics is the study of how humans perceive sound, and it's relevant here because advocates of lossy data compression argue that when listening to CD-quality audio, it is impossible for our brains to perceive all the data reaching our ears. It is, therefore, unnecessary — the argument goes — to store and reproduce all of that data. But which data can be removed is another question, and this is why various psychoacoustic principles are exploited in different amounts by different perceptual audio coding algorithms.
I wonder if different people will perceive sound differently (in terms of frequency and dynamics) purely based on physiological/psychological traits?
Not completely sure, if this is true, I couldn't find the study for it right away.
A bit that bothered me was the description of how the encoders work. It mentions the discarding of frequencies, which is not entirely correct. The only real discarding of frequencies happens with the lowpass filter that is typically applied as a pre-processing step, not the actual encoding. The encoding of the frequency components happens with coefficients, which are quantized (coarsely encoded), not zeroed. This introduces noise to the signal, which is weighted against the computed masking threshold, driving bit allocation to the different frequency bands.
As for the format recommendation for people ripping their own CDs, I'd say stick to a lossless format for storage at home. That way you're future-proof. For mobile use, make encodings from those files using whatever best lossy format is available at the time.
The point of lossless audio compression is archiving - knowing that you can make a first-generation encoding of the audio you have stored for any new device or format in future.
If you want to listen to your FLAC files, great, you're not losing anything by doing that - just don't try and tell me that it's an aesthetic choice that makes the slightest bit of difference to the experience.
It's just basic human nature. As soon as you say "you can't see/hear/taste/perceive the difference between X and Y" some dick-waver will come in and start saying they can do it. And of course, the first to admit they can't hear a difference loses, and then you end up with things like Monster Cables, and the Emperor's new hi-fi setup.
I've seen people seriously claim they can hear the difference between the same audio signal burned onto different brands of CD-Rs.
[0](https://forceincmilleplateaux.bandcamp.com/album/most-beauti...)
https://benchmarkmedia.com/blogs/application_notes/intersamp...