Neil Young on digital audio: You're doing it wrong

Install foobar2000 & foo_abx component on your PC, get some Hi-Rez samples, convert them to 16/44 using fb2k's Converter utility (with linear-phase SoX Resampler) & fire away!.. Having a decent DAC (say, a 24/96 one) may help, too.

 

When Neil is comparing the music in his car it is all coming from one file. His iPad allows him to switch the output at anytime whilst the music is playing. For example, he may have it on 24/192 (from a 24/192 master) and then switch to 16/44.1 to demonstrate the difference.

Some of the testimonials have mentioned Neil doing this, so all demonstrations are sourced from the same high resolution file. Most of the time it's Neil's own music that he uses. I think he has said in an interview or in a magazine article recently that he has remastered his whole back-catalogue at 24/192.

 

Unless psychological bias can be filtered out, "testimonials" are rather meaningless. Another example is eyewitness testimony (regarding criminal proceedings). It is notoriously unreliable.

 

I'm fed up with arguing! It's pretty clear to me through a lot of personal experience (that most definitely isn't my imagination) that high resolution music gives you a wider stereo width, more ambience and most high resolution releases are a lot less compressed when it comes to dynamic range, although they could do the latter for CD as well.

When converted to 16/44.1 for CD, this effect that I've just mentioned is lost due to the fact that you're only hearing the master tape 44, 100 times per second, which actually isn't that much. I'm talking mainly about music taken from analogue tape. High resolution allows you to hear what the studio master sounds like, or similar if it's tweeked a bit during the mastering process for a high resolution release. I dare say that even in the world of high resolution, certain master tapes may benefit being EQd slightly, although you will still hear the full master tape quality.

The CD is incapable of giving you studio master quality of a 2-track analogue tape. It seems strange that people on here seem to refuse to believe that high resolution audio is better but when it comes the old vinyl vs CD debate... the opinions are far more even.

The negative elements of the CD were noticed as far back as 1993 by numerous people in the music business who noticed that something was missing when comparing CD to vinyl. I've already posted this video once but I'll post it again in case anybody missed it. The idea of a higher sample rate to overcome this problem was mentioned in this report...

 

The negative elements were noticed on debut in 1982. Fortunately, sound reproduction from CD has improved vastly from those early days. In most cases, the problems were with the players and not the A-D or the discs themselves - I admire some of my early CDs greatly now. Nevertheless, I prefer SACD.

 

So you're saying that you will be able to hear the difference between 24/176 high resolution sample & its 16/44 incarnation when the delta between the two is:

- the waveform:


- the spectrogram:


- the spectrum of the highlighted in the previous 2 graphs excerpt:


I seriously doubt that (unless you have bat ears, of course)...

 

Yes.

My ears simply don't hear the difference.

 

Yes. I too did a ABX test and couldn't pass. There is a ton of discussions on this topic over at hydrogenaudio.org and on other sites, and not a single person could ever present clear and verifiable positive results of their tests being able to hear the difference between 16/44.1 and 24/192 or other hires. Not a single one. Of course the test has to be done with proper samples, it must be exactly the same recording, properly down-sampled and dithered to 16/44.1.

 

It's 2014. Why on earth would anybody want to choose "believing" over scientific proof achieved by proven scientific methods??? Especially on a subject that's so easy to understand and perform experiments.

 

Sorry, but you seemingly have no idea how digital sound works. You do not hear anything "44,100 per second". You hear a continuous analog sound wave going from your speakers or headphones into your ears, the same analog sound wave as you hear when you play vinyl or any other analog medium. That analog wave is generated by the reconstruction filter in your DAC.

 

Right, like there's no expectation bias when your good buddy Neil Young gets you lightly toasted while evangelizing about his awesome idea, sets you in his fancy car, and cranks music with a big smile on his face. I'm sure everyone who listened did so with a robotic, emotionless, critical ear.

Also, 32 bit DACs are a fricking joke. Don't fool yourself; there is no sonic benefit. I look forward to the next big thing, the 64 bit DAC, or even 128! If you want random digits on bits beyond 22 (and that's being charitable) enjoy them but they aren't contributing to the sound.

Digital audio to Neil Young: You're doing it wrong.

 

I think there are several reasons why people come to the conclusions they do.

Not all are always based on the realities of what is actually better or not.

Human bias, expectations, heck even just plain excitement and hope factor in.

Until all those built in human "emotions and thoughts" can be removed from the equation, there will always be a chance of a skewed response.

Until we can be absolutely sure, the same "Exact" mastering is used, we can never simply compare and get a reliable answer.

I think "Higher Res", has some benefits, but honestly ( not from lack of hearing, bias or lacking equipment) I find it really hard to say for sure, if those benefits are always due to simply a deeper noise floor AND extended High frequencies ONLY.

Some changes I thought I heard, seemed to come partially at least, from the domain of "Mastering type" alterations.

I am left thinking it might be beneficial, but just not totally sure though.

 

Ahh, but it's not the increased range that is getting me excited about 32 bit dacs, it's the lower floor noise and better timing which are making all digital sound that much better in play back regardless of file size. The Chord Hugo does just that. Here is a posting from the creator himself from last night, Rob Watts.

"I am constantly amazed at how good AAC sounds - much of my listening is BBC radio 3 on iplayer - the through the night program. I just turn it on, work, and every so often I hear something that catches my breath. And this is with 192/128 kbs AAC+. It is very puzzling that the timing benefits of Hugo also very much works with compressed music.The compression algorithms have coding specifically to preserve transients, and it works. Hugo is a format leveler, in that I have heard truly engaging emotional music from AAC, CD, and DSD. The big benefit of Hugo is not it's sound - it just connects you to the music. "

It's an exciting time in Digital Technology!!

 

Well said, Kevin. I used to fall prey to that kind of thing, but now I question everything and demand results rather than expect them.

 

So the information in that video I posted was wrong?

 

Yes.

The whole "digital audio is slices, steps, is non-continuuous, is missing information, rough approximation, etc..." talk is complete bu115hit. It's lies and propaganda coming from a total misunderstanding how digital audio works:

http://wiki.hydrogenaudio.org/index.php?title=Myths_(Vinyl)#Vinyl_is_better_than_digital_because_the_analog_signal_on_the_vinyl_tracks_the_analog_signal_exactly.2C_while_digital_is_quantized_into_steps

 

The graphic in the video showed "stair steps" for digital audio which is not scientifically accurate.

While the video below is considered heretical by some (even here in the forum), it is solidly rooted in science and explains scientifically (as opposed to the pseudoscience that Michael Fremer used in that 1993 MTV video) how the A->D and D->A process works. I encourage you to watch it and (if possible) watch it somewhere that has decent audio capabilities so you can adjust the volume to hear the dither noise.

No doubt Monty lacks reverence for certain "audiophile beliefs" and that lack of reverence earns him scorn from some in the (and this) community. But I have yet to see any counter arguments to his points that are rooted in anything other than those aforementioned "audiophile beliefs".

 

Man, I could hear that dither noise from outside with just my crappy computer speakers playing! And the difference between 16 bit dither and 24 bit dither is HUGE!!!

 

If you're talking about the video, I think you mean the difference between 8 bit and 16 bit.

 

This is just not true, it is not how sound works (or DACs)

We don't hear individual samples. An individual sample can be part of almost any frequency.

We hear samples in context of other samples, or more precisely, the frequency and amplitude that results from the sinusoidal wave form generated by other samples before and after it.

Unless the DAC is malfunctioning, every frequency nyquist and below is reproduced by the DAC. With the amplitude of the samples, there is quantization error - and there, bit depth helps. The higher the bit depth the lower the quantization error. At what point the quantization error is too quiet to hear is the question. It certainly is with 24-bit audio and I believe it is with a properly mastered 16-bit audio at safe listening levels.

The sample rate, the only impact sample rate has is the bandwidth - what frequencies can be reproduced. Higher sample rates simply do not do a better job at reproducing frequencies within the bandwidth range of lower sample rates.

It is frequencies that determine what pitch / tones we hear and it is amplitude that determines how loud we hear them.

Higher bit depth does mean better resolution / lower quantization error. Higher sample rate just means more data and more possible bandwidth, not better reproduction.

It does not mean you are listening to the tape X times per second.

With just 3 points, I can make a perfect circle that passes through all of them. Having 27 points doesn't let me draw a better circle, just means there is more data - not better results.

The same is true with the reproduction of sinusoidal waves. A sample rate of X lets you reproduce every frequency X/2 and below. Sample rates above X does not do a better job at frequencies below X/2 unless my lowpass filter isn't working and the X sample rate is trying to describe frequencies above X/2.

The sampling theorem was derived independently by many different mathematicians / scientists and has been proven using several different approaches. It is very solid mathematics.

 

Hopefully Andrew is too busy running his Foobar tests to reply.

 

N.B. with modern freeware audio apps (like, foobar2000, Audacity, etc.) the non-dithered 24-to-16 bit conversion appears to be done using rounding (not truncation) and the 17-th bit of of the 24-bit signed integer value's is not discarded. Thus, quantization errors with no dither applied peak are at -96.3dBFS (spectrum averages at ~-134dB), not at -90.3dBFS, like with truncation. Using non-intrusive noise-shaped dither (I use MDA dither VST plug-in currently) can help drop spectrum of the quantization error to -140dB (and below) in frequency range, where human hearing is most sensitive (mids). But even with no dither applied 24-to-16 bit modern (i.e. rounding) conversion is pretty good, compared to good old truncation...

 

I still think, the actual recordings and all that come before what we listen to, matters a lot more, as far as determining if something sounds good or not.

I get why "higher-res" could or should be better, but in about 70-80 percent of recordings, I feel like it is trying to put a spoiler on a Pinto, and claim it makes it handle better at high speeds.

 

For those interested here are the spectra of 24-to-16bit quantization error at 44.1 kHz sampling rate for AUdacity (RectAngular, TriAngular & Noise-Shaped dithers), FB2K's native dither & MDA dither VST plug-in:



As you may see, rectangular & triangular dithers in Audacity are rendered pretty much useless by rounding (NO DITHER graph)... Audacity's & FB2K's native noise-shaped dithers are really good, but produce rather high noise peaks (-70...-60dBFS) for my taste, YMMV. Both Audacity & FB2K are running in 32-bit float mode.

 

I dither used SoX shaped dither (dither -s option) when resampling with 16-bit output. I don't bother dithering with 24-bit output.
I'm not sure where SoX shaped dither would fall on that graph, but quantization noise at 16 bits is never something that bothered me, I don't play music loud enough to hear it.