Neil Young speaks out against excessive compression

[DWBass]

Ok, so maybe I've been comparing mp3's of old. Does anyone know how long this quality of mp3 has been around? I suppose if the quality of mp3's has increased in say, the past 2 or maybe 3 years (3's probably a stretch) then I may have to re-evaluate.

In my opinion...the quality of a high bitrate mp3 (or any lossy format) is hugely better than in years past! And I think some folks are still referencing those old encoded files. The whole point of the continuing evolution of the encoding programs & process is to keep the audible parts as close to (audibly, not actual) cd quality as it possibly can in a smaller file size. It's only going to get better and better!

Lastly, a good % of cd's are compressed right out the box so even though it's a cd, it's still not a fully reproduced rendition of what was recorded. There's not a lot of grass roots recordings out there!

I'd also like to say, it may be the listening system (outside of portable players) used that could be causing a discerning ear for some!

 

[swspiers]

There will ( most likely) never be one 'smoking gun' study, example, test, clinical trial or anything that will put this debate to rest.

There is a plethora of information on psychoacoustics. The book 'This is Your Brain on Music' is a great example and a good place to start. One of the reaons that mp3's work is that the brain actually fills in the gaps of missing information. Kinda how most amps and cabs don't produce the low B fundamental on a 5-string bass, yet he 'hear' it. We're actually hearing the harmonic @ 60 Hz. And because it's the second harmonic, the brain identifies it, unconsciously, as low B (30 Hz) This ability to fill in the gaps is what accounts for listening fatigue.

Is there an immediate, audible difference? Not that much.

Is there a perceivable difference over time? yep.

So the subjects may not be able to tell the difference between mp3 and full resolution per se immediately, but over time they can determine a preference.

 

[Febs]

There will ( most likely) never be one 'smoking gun' study, example, test, clinical trial or anything that will put this debate to rest.

This may be true with respect to things like cables, speakers, maple v. rosewood, etc., because with all of those things, there are some significant logistical challenges to conducting a proper double-blind test which make such tests impractical or expensive. That is not true of audio codecs, where anyone who claims he can hear a difference between a lossy-compressed file and the uncompressed original can provide proof of that assertion quickly and for free, if in fact that assertion is true.

 

[Queg]

And then ask them to do it again in a blind test using well encoded mp3s and they won't be able to tell the difference.

So until we conduct blind testing on every claim of sensory preference, we must assume the preference is merely placebo effect?

Here's the blind test I'd like to do, a test that ultimately addresses what matters when it comes to music: PROLONGED ENJOYMENT

1. Provide 2 blind copies of a good recording of acoustic music, one being MP3 the other HD digital..
2. Provide a decent but not extravagant playback system, say a quality DAC, good headphone pre and Sennheiser HD650's.
3. Require the test subject to listen to the recording at least a few times a day for a week, with a 2-day initial requirement of listening to both recordings, followed by the option of choosing recording "a" or "b"
4. Compile results. Did they develop a "blind" preference for one recording over the other?

It's not about soundbites and rushed A/B/X tests, or how fast you can download it, or whether it not it sounds good enough on your headphones at the gym, what matters are those elements that lubricates your potential to actually sit down and be drawn into the music.

 

[Febs]

So until we conduct blind testing on every claim of sensory preference, we must assume the preference is merely placebo effect?

If you are going to assert that there is a difference that a signficant number of other people cannot perceive, yes.

It's not about soundbites and rushed A/B/X tests

Red herring. You can conduct the ABX test with full songs, and listen for as long as you like. There is no need for it to be limited to "sound clips" or be "rushed." The results do have to be repeatable over a statistically significant number of trials, however. Your proposed test would not yield any conclusive results.

 

[lomo]

I understand the inherent inability to truly evaluate vinyl due to the nonmusical artifacts. As for the digital evaluation, we each played for the other, who was blinded, the same clip wich was saved from CD to iTunes using MP3, Apple lossless and WAV methods. We did this for 10 sample from different songs and compared only 2 filetypes at a time. We could both differentiate the MP3 from WAV 7 or 8 times out of 10, and both failed to distinguish lossless from wav. The system is a Musical Fidelity integrated amp, Meadowlark speakers and Macbook through an upgraded, outboard DAC.

Obviously this was not a large sample, but I found the results quite compelling. I am definitely NOT a "true believer" in any of my worldviews, as I try approach everything with skepticism
and do not choose to believe that which requires the suspension of reason

Telling vinyl from CD can be easy for reasons independent of the limitations of the medium. A record spinning will have "tells" like low frequency rumble, scratches, etc. That and the fact that the master used to source a record is virtually ALWAYS different than the one used for a CD, if for no other reason than to accomodate the RIAA curve.

CD from MP3 tests - what was your methodology? What was the encoder and bitrate used? What music did you select? These are all important variables to mention.

 

[Febs]

I understand the inherent inability to truly evaluate vinyl due to the nonmusical artifacts. As for the digital evaluation, we each played for the other, who was blinded, the same clip wich was saved from CD to iTunes using MP3, Apple lossless and WAV methods. We did this for 10 sample from different songs and compared only 2 filetypes at a time. We could both differentiate the MP3 from WAV 7 or 8 times out of 10, and both failed to distinguish lossless from wav.

Did you normalize the clips to ensure that they were the same volume?

 

[Alkali Ike]

Respectfully, anyone who can't hear a difference between AAC-256 and even just Red Book CD is in denial, or has hearing loss, or their playback system is not up to snuff. When I rip CDs in iTunes, I make 2 copies: One in Apple Lossless, and one in MP3-320, the highest bitrate MP3 possible. The MP3 files sound pretty good and are small enough for portable pleyers. Anything less than 320 and I can tell an OBVIOUS loss of quality.

BTW I use MP3 format because my car's audio player supports it. For home playback I use the Apple Lossless.

 

[moles]

There actually ARE blind tests that back up the notion that humans can't discern Redbook CD audio from "hi res" and SACD formats. I will post the link tonight or tomorrow morning.

Then there's the fact that many labels use DIFFERENT MASTERS for their SACD layer, rendering the comparison moot.

Third reference to SACD that I've read in this thread (so far...still playing catch up on this today...)
You do all know that SACD failed because it was all BS, right? There is a formula to convert the 16bit/44.1kHz data stream to a single bit stream, and SACD falls EXTREMELY short of the number required to match Redbook, never mind exceed it.
Furthermore, the way they tried to get around it is to actually replace the content above a certain frequency threshold with a white noise, triggered by the player AFAIK. Hardly what I'd call hi-rez.
Add to this the speculation that the labels/manufacturers actually downsampled, squashed with compression, or otherwise treated the CD layer included on SACD, in order to present the SACD layer as "sounding better"...
I'll post something later to back this up, of course - but I'm late for a meeting right now and have to look through the bookshelves for 8 year old course notes to be able to post actual numbers.
Anyway just wanted to point out that the SACD failure isn't really a fair argument for consumers not caring about hi fidelity, since it may be the lack of that fidelity that actually killed the format.

 

[lomo]

Did you normalize the clips to ensure that they were the same volume?

We adjusted the volumes by ear to be roughly equal, but as I said, this was not an exhaustive, rigorous exercise. I honestly did not expect either of us to be able to differentiate, let alone both of us. Obviously, the sample was small and the procedure single-blinded, but it is interesting (to me) nonetheless.

 

[projectMalamute]

Did you normalize the clips to ensure that they were the same volume?

You can't do that. You get dynamic range compression with any of these lossy codecs, that's part of how the psychoacoustics work. If you peak normalize the compressed file will be louder than the uncompressed.

 

[DWBass]

Respectfully, anyone who can't hear a difference between AAC-256 and even just Red Book CD is in denial, or has hearing loss, or their playback system is not up to snuff. When I rip CDs in iTunes, I make 2 copies: One in Apple Lossless, and one in MP3-320, the highest bitrate MP3 possible. The MP3 files sound pretty good and are small enough for portable pleyers. Anything less than 320 and I can tell an OBVIOUS loss of quality.

BTW I use MP3 format because my car's audio player supports it. For home playback I use the Apple Lossless.

Wondering what type of system you are listening with at home?

 

[Queg]

If you are going to assert that there is a difference that a signficant number of other people cannot perceive, yes.



Red herring. You can conduct the ABX test with full songs, and listen for as long as you like. There is no need for it to be limited to "sound clips" or be "rushed." The results do have to be repeatable over a statistically significant number of trials, however. Your proposed test would not yield any conclusive results.

Curious, why would my proposed test, which could be easily repeatable and involve a statistically significant number of trials, be any less conclusive than a a typical ABX test, which despite your claim of red herring, are usually conducted in constrained, time limited environments?

 

[fishdreams]

The difference between cd and vinyl lp is palpable, especially in the bass. If you for example buy this record on lp Soul Jazz Records – New Orleans Funk – The Original Sound of Funk 1960-75 you will hear that George Porter's tone is way huge on lp, and certainly of a different quality. (I have both a copied cd and the actual lp). Apples and oranges, really.

 

[projectMalamute]

Given the rates data can be transmitted and the dwindling cost of storage why would we want to use the most marginal system we can possibly get away with?

If even only one in 20 people can reliably tell the difference, so what? Whatever happened to excellence?

 

[lomo]

Given the rates data can be transmitted and the dwindling cost of storage why would we want to use the most marginal system we can possibly get away with?

If even only one in 20 people can reliably tell the difference, so what? Whatever happened to excellence?

Cost. Excellence in most aspects of life is available to those willing and/or able to pay extra for it in some way (money, time, effort, etc).

 

[Febs]

You can't do that. You get dynamic range compression with any of these lossy codecs, that's part of how the psychoacoustics work. If you peak normalize the compressed file will be louder than the uncompressed.

While that is contrary to my understanding, I admit that it has been several years since I really did any reading on the subject. I know that Foobar's ABX comparator, for example, allows you to do some sort of volume adjustment, although come to think of it, it may be something more along the lines of ReplayGain than peak normalization.

Curious, why would my proposed test, which could be easily repeatable and involve a statistically significant number of trials, be any less conclusive than a a typical ABX test, which despite your claim of red herring, are usually conducted in constrained, time limited environments?

If I'm understanding your test correctly, the listener always knows whether he is listening to A or B. Now if you randomized A and B for each trial so that the listener didn't know which recording he was listening to, and he consistently chose one or the other a statistically significant number of times, I would be more inclined to agree with you that the results conclusively indicated that he was able to tell a difference. If that is what you're suggesting, then we are not too far apart conceptually, although I still prefer an ABX test because it tests only whether the listener can identify a difference, without requiring the listener to make a subjective determination as to which he prefers.

 

[SteveC]

What we need is music to be released in 32 tracks so that everybody can do their own mix. And a few digital automation tracks so that "producer" mix can be heard.
Want to hear just the bass? Just turn on the bass track.

Yes. I'm sure there could be an app for that.

 

[eqvolvorama]

Any of you no-discernible-difference-absolutists read this study? I say eat your words, suckas!

Ok now I'm just a troll and need to get a life...maybe I'll go pop in a cd or throw a record on the turntable!

..no hard feelings, seriously.

The study compares 128kbps MP3s to Lossless, using a 2-year-old encoder.

The result is not terribly surprising, but as I read the study, it's notable that at 128kbs, a bitrate nobody really uses anymore, a considerable number of young ears still got the study "wrong." My hypothesis would be that if you upped the ante to 256kbps, the results would be much different.

The neutral loudspeakers thing is an entirely matter altogether. Some loudspeakers are designed to impose an EQ curve. Some are PAINSTAKINGLY designed to be as flat as possible. There are audio differences between the former, typically consumer grade speakers, and the boutique speakers designed as studio monitors.

 

[eqvolvorama]

There will ( most likely) never be one 'smoking gun' study, example, test, clinical trial or anything that will put this debate to rest.

There is a plethora of information on psychoacoustics. The book 'This is Your Brain on Music' is a great example and a good place to start. One of the reaons that mp3's work is that the brain actually fills in the gaps of missing information. Kinda how most amps and cabs don't produce the low B fundamental on a 5-string bass, yet he 'hear' it. We're actually hearing the harmonic @ 60 Hz. And because it's the second harmonic, the brain identifies it, unconsciously, as low B (30 Hz) This ability to fill in the gaps is what accounts for listening fatigue.

Is there an immediate, audible difference? Not that much.

Is there a perceivable difference over time? yep.

So the subjects may not be able to tell the difference between mp3 and full resolution per se immediately, but over time they can determine a preference.

An analogy of what compressed lossy audio does is as follows:

Let's say you are having a conversation with your friend at at a table. You are speaking at a normal, audible level.

Then someone walks by and turns on a vacuum cleaner. The sound of your friend's voice will be as loud as before, but in your head it will be as if they've been switched off. All you will hear will be the vacuum.

When the vacuum cleaner is turned off, you will hear your friend again.

The job of the lossy encoder is to not bother to encode the sound of your friend speaking when the vacuum cleaner is on.

That said, the "perceivable difference over time" thing has yet to be demonstrated outside of placebo effect, as far as I know. It's something people suspect, but even when they've been trained to listen to an MP3 of a certain song for a long time, I haven't seen any studies which indicate they'd be any more prepared to pass a blind test against the lossless source.