Surprise: Neil Young still hates digital music Neil Young, who a few years back famously described Apple as the “Fisher-Price” of sound quality, is giving his “I hate digital music” can another kick, claiming that even the late Steve Jobs listened to vinyl rather than his own company’s inventions. Speaking to Peter Kafka and Walt Mossberg at the “Dive Into Media” …
Listening to overcompressed music is like trying to look at a picture while someone shines a torch into your eyes.
Current mixes and remasters of "old" songs invariably strive to bring the peak level of every instrument up to the max and to bring the dynamic range down cramming it all into one noisy incomprehensible mess. Try listening to a decent quality CD from the 80s and then compare it to a remastered recording. The artistry of having subtle instrument sounds is lost as they are all brought up to be right in your face. The objective seems to be to make it subjectively as loud as possible, but perceived loudness is at the expense of loss of detail. Then it gets compressed further to mp3, and if you buy that format you can't get back the bits that were lost in the conversion to mp3.
I bought Bingo! by the Steve Miller band (no accounting for taste I know) on CD and it has to be the worst mix I have ever heard. I played it and never got past about a minute of each track before skipping to the next. I like the music but can't bear the recording/mix. It sounds like every channel on the mixer is cranked up to 11, and even when played quietly it is cringeworthy.
However, if you can't hear the difference I suppose you don't care. The market long ago stopped striving for excellence and settled instead on providing the minimum quality they can get away with that the majority of people will accept. A cynic might think this was so that later you can charge them again to buy the same thing in a lossless format.
I think the worst one I've heard so far is 'This Gift' by Sons & Daughters. It's physically uncomfortable to listen to for more than 1 track at a time, and even that is unpleasant.
I've spent a lot of money getting hold of original non-remastered versions of other albums that I'd sold in the past. Here's a comparison of the original and remaster of 'Money For Nothing' by Dire Straits:
http://www.youtube.com/watch?v=7UjQc0dM4H4
I've compared the actual CDs on a decent system, it's like night and day. Most of this seems to be driven by catering to people with rubbish playback devices (iPods+etc with stock earbuds) - perhaps they need to start making two mixes of everything, a standard and portable one with higher volume at the expense of range.
The big problem with all this is not that people in noisy cars or with iPods can get access to music that sounds sort of OK on their equipment, it's that only having that mix available means good quality recordings of some recent music simply don't exist at all. The music is forever unlistenable, you can't fix range compression.
Eric Clapton, 461 Ocean Boulevard, released 1974, copyright on that would last 50 years and would have expired in 2024, remastered deluxe[1] version released in 2004, so copyright would expire in 2054.
Oh look, by clever slight of hand, the copyright mafiaa get an additional 30 years copyright.
[1] deluxe version: copyright mafiaa marketing speak for "we've added the really shit tracks that weren’t good enough for the original record"
You don't need ultra-high "resolution" better known as sampling rate, to capture the frequencies available to the human ear, see http://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem
Also there are many music stores, eg 7Digital who run the Ubuntu one music store, that support FLAC audio, an open lossless audio format - a new format/device does not need to be developed: just submit patches if you think you can do a better job.
The Sampling Theory applies to continuous signals, rather different from the complex shapes of musical notes which have attack and decay - not just a bunch of frequencies, but contained within an 'evelope' shape.
There is also a practical problem with replay of digital audio streams, which is jitter - imperfections in the clock frequency of the digital stream.
Another lesser problem with signals resulting from D->A conversion is the sampling noise at frequencies above 22.1kHz - while inaudible to adults it may affect the performance of equipment that was designed on the assumption that the input signal contains only audible frequencies. Example: some metal-dome tweeters have resonances around 22-25 kHz.
But these three effects pale into insignificance when compared with the destruction wreaked by modern dynamic range compression.
Although the technical performance of CD is better than vinyl, the debate still rages - I think because the shortcomings of vinyl reproduction are less noticeable when concentrating on music. Perhaps because static click are essentially random, and surface noise is concentrated in the lower frequency range and therefore less objectionable.
Most unkind, but he has a few valid points further on.
Shannon/Nyquist is a bit more complicated than finding one's way to the loo in the dark (at least it was for me). >;-)
I've seen people come off the rails early on with this stuff when they fail to comprehend that a comparatively regular signal represented in the time domain [normal oscilloscope-type display] can contain many complex harmonics. Understanding the mechanics of Fourier maths is one thing, but for some, picturing in their heads how all those signals are represented by a single trace or graph line is another matter altogether.
If you think about it carefully, it's far from intuitive, especially so when one is simultaneously trying to imagine the same signal in the frequency domain by conceptually looking at right angles head-on down the time domain axis.
The next hurdle is to understand is why the sampling rate is usually double plus a bit of that of f(max) for simple streams, audio etc. To test someone's understanding of the Nyquist-limit concept, he/she should then be able to explain why the sampling rate of some data streams (certain types of skewed images for example) might ideally need to be three or more times that of f(max).
Of course, that's just the beginning; stochastic processes/probability theory etc. etc. have been known to produce bad hair days in some, yours truly included.
I was being unkind quite specifically, and without that basic understanding some of the latter points are not quite comprehensible. Other comments following mine that address some of the more interesting aspects of what we actually hear as opposed to what a simple model (aka frequency limited model) of what we should perceive are very relevant. But we do need to understand the basics first :-)
> Another lesser problem with signals resulting from D->A conversion is the sampling noise at frequencies above 22.1kHz - while inaudible to adults it may affect the performance of equipment that was designed on the
That is a problem, but the frequencies above 22Khz are not passed on to the amplifier (on any decent equipment) the problem is that the level of filtering required causes phase distortions in the frequencies you want to keep. This is solved by digital oversampling, shifting the unwanted frequencies to 98 KHz and above.
> But these three effects pale into insignificance when compared with the destruction wreaked by modern dynamic range compression.
Yes.
> Although the technical performance of CD is better than vinyl, the debate still rages
This will probably be down the fact the majority of harmonic distortion in a vinyl system is even harmonics, and therefore musically related to the original frequency (i.e. the same note an octave higher, for example), D-A distortion is more odd order harmonics, which just sound wrong. (Which is also true for valve vs. transistor).
However I doubt this is an issue in any decent quality D-A from the last decade or newer.
Also on a decent vinyl system pops and clicks actually have a position in the stereo sound field and can be ignored just as a noise in the room can.
That's true if you start from the false premise that only the range from 0 to 20 kHz is audible.
The 20 kHz figure for hearing response is an average derived from the perception of simple tones emitted from speakers outside the ear canal. Age, health, genetics and general variation within the population mean that some people can hear above this. In some experiments, with emitters placed against the bones of the skull rather than outside the ear, people could detect signals up to 100kHz.
But even assuming that everyone's hearing is limited at 20 kHz, you need to consider that your hearing doesn't behave exactly like a Shannon/Nyquist system. Consider the effect of beat frequencies where the two fundamentals are above the nominal range of hearing, but their beat product is audible. On a conventional A/D process, these are stripped by the initial antialiasing filter, and thus lost, so the easiest solution is to up the sampling rate. Doing this also prevents any uneven response from the antialiasing filter allowing alias frequencies into the processing chain (where they will interfere with the high-frequency information you want to capture).
Sampling at a higher rate also allows greater phase accuracy, and reduces the negative effects of off-phase signals when sampled at low numbers of samples per cycle (as you approach the Nyquist Frequency, the phase on an input tone causes a change in the resulting amplitude. Worst case is at f=0.5fs, where a 90 degree lead/lag results in no output signal at all). Phase of signals is also important for placing sounds in a stereo sound-stage, and the most significant part of the spectrum for this is the high frequencies.
There's lots of reasons for going to much higher resolution audio, and once you do, of course you can pack it with FLAC, because the format imposes very few limits on frequency or sample resolution; but the FLAC file format alone can't solve the original problem of not recording the signal accurately enough.
100KHz ... that's RF. I wonder if these subjects could actually "hear" the signal as opposed to psycho-acoustically experience it as some other sensation via induction of the signal into the auditory/facial nerve(s).
I'd also imagine that, mechanically, the eardrum couldn't vibrate at 100KHz (not even in young chilodren) and what they were hearing was perhaps a sub-harmonic that fell in the normal auditory range.
I don't know so do correct me if I'm just spouting nonsense.
But I did find your comment interesting and thought-provoking.
Most far-out theory is that it may have even been an inductive coupling from the speaker coil directly to the subject's brain.. hmm, that's how my induction hob works, so could be something in that.
... I just remember reading about it once and the research itself was pretty sceptical about whether the subjects were actually "hearing" the thing at all, but they were definitely perceiving it.
This post has been deleted by its author
Historically, we've ended up with 44.1 and 48kHz sampling rates as part of standards, now we're forever trying to get around their apparent limitations (although in double-blind tests I've never conclusively picked them as noticeably inferior to say 96kHz, nor could any of my colleagues who adamantly considered themselves as members of the Golden-Ear Brigade).
Clearly, it would have been better to have moved the 'Nyquist-limit problem' down into the analog domain and then let one's ears do the integrating (but when the standards were set this would have been considered an extravagant waste of bandwidth and storage).
What I mean by that somewhat strange comment is that had the baseband been 0 to 40kHz and the sample rate 96kHz then phase issues with clocking/sampling would be insignificant even though the 2:1 ratio would have been retained (remained the same). Scaling up to 196kHz would then essentially achieve nothing.
It could be argued that a better approach than increasing the upper baseband frequency limit to 40kHz would probably have been to increase the depth of the sampling from 16 to 18 or 20 bits (dynamic range limitations are unquestionably audible under good conditions). I'd certainly favour this approach over increasing the upper recordable frequency limit.
Perhaps a theoretical case could be made for fixing phase errors when employing a 40kHz baseband by using a 4x f(max) sampling frequency but I'd require some rather strong evidence before I'd be convinced.
This post has been deleted by its author
I remember reading that Philips wanted to emply 14-bit sampling mainly because they had a line of 14-bit DAC chips. Sony, the other partner in the venture, insisted on 16-bit sampling even though the converters at the time struggled to produce anything meaningful below 14-bit resolution, in order to future-proof the system "into the 21st century" (not a bad prediction)
The 44.1k limit was imposed by mastering requirements. The only readily-available recording medium with the bandwidth required to store the digital bitstream was U-matic broadcast videotape. Unfortunatley, the record and playback equipment, being designed for broadcast use, inserted blanking signals at the end of every "scanline". However, 44,100 samples at 16 bits per sample plus error correction, allowed maximum use of each scanline period on NTSC players, and so a "standard" was born.
Later formats, freed of the requirement to be compatible with equipment designed for a completely different purpose, used 12/24/48k as their sample rates, because it pushed the "near-Nyquist" problems higher out of the audible range. This is why DAT, DVD and BluRay use 48, 96 and 192k sampling respectively.
Actually, here's a very interesting article explains this history, first hand, and better than I can: http://www.exp-math.uni-essen.de/~immink/pdf/beethoven.htm
The last Neil Young album I bought, probably about 25 years ago now, sounded as if it was recorded in a garage on an old Grundig using a crystal microphone. I don't think higher resolution audio would have done any good, especially given the musical skill of his Wyld Stallions - uh, I mean Crazy Horse - backing band. It's the only music CD I ever threw away.
Trouble is, that "rawness" makes his stuff some of the best going for shining an umpty-million candlepower searchlight on the limitations of audio playback systems. Many of the things that make a reasonable fist of autotuned music are utterly shit at harmonics, fingers squeaking on strings, foot shuffling, breath sounds and passing aircraft.
Oh and it's "Wyld Stallyns" BTW....
This article in El Reg last November does a great job of diving into the technical nitty gritty of Neal Young's grievances. Worst thingis that things like widespread ignoring de-emphasis flags gets people used a particular forms of distortion to the point that audio as Neal intended might "sound funny".
http://www.theregister.co.uk/2011/11/24/digital_audio_history_part_one/
Sample rate is the number of "slices" of the music you can take in any quantum of time. But that's only 1/2 (at best) of the "resolution" (perceived quality) equation.
Another key part is by how much each slice can vary from the slices taken either side of it and the difference between the minima and maxima of each slice - the dynamic range.
Besides which, the sampling theorem you link to relies on interpolation to reconstruct the "original signal". But "approximate a facsimile of" would be more accurate than "reconstruct". If there is some subtlety in the original that the interpolated reconstruction glosses over then you haven't reconstructed the original at all.
Now, whether you notice the difference... that's a whole different question. But saying that just because you can't notice the difference there is no difference is just flat out wrong.
Ask any digital photographer about the difference between RAW and JPEG for example... they will tell you there is a significant difference. But ask someone just looking at a RAW image on screen vs a JPEG of the same RAW, and they will say there is no difference.
Okay, so the human ear (a good one) has a dynamic range of approximately 140dB correct?
IIRC DVD audio is defined as being 24-bit linear PCM or AC3; taking the former case, that means there are 2/2²⁴ approximately different amplitudes that can be represented:
Python 2.7.1 (r271:86832, Sep 3 2011, 03:15:40)
[GCC 4.4.5] on linux3
Type "help", "copyright", "credits" or "license" for more information.
>>> import math
>>> levels = 2.0 / (2**24)
>>> range = 10*math.log10(levels)
>>> range
-69.23689900271567
>>> range = 20*math.log10(levels)
>>> range
-138.47379800543135
Bloody close don't you think?
This post has been deleted by its author
"A CD can construct the original wave. 16 bits and 44.1 KHz is enough for perfect reproduction of a band limited signal."
You are just plain wrong. You first convert a continuous wave into discrete numbers - you get quantisation noise. The lower your sampling rate, the lower the signal to quantisation noise ratio will be. That's why they have to add dither to CD signal after AD conversion.
Next, you need to reconstruct a continuous wave from discrete numbers - so your DAC must interpolate between samples. The lower the sampling rate, the higher the interpolation noise.
I think you over-estimate the hearing capabilities of the bulk of the population; if they could tell the difference then they wouldn't have switched from portable CD players to MP3 players in the first place.
My father-in-law genuinely can't tell the difference between Metallica and Paloma Faith, as far as he's concerned they live in a box labelled "stuff I don't like". It/'s the same story with the majority of my friends, they categorise music as "good" and "rubbish", mostly based on whether they encountered it before or after they turned 20 or if they identify with the image of the artist or not.
It saddens me greatly, but I think you'll find they are the majority and that they can't tell the difference. If they could, then Simon Cowell would be a pauper.
Download foobar2000 (a music player), then the ABX comparison plugin.
Get one of your CDs and rip to WAV.
Now convert that WAV to a 256 kbps MP3 using EAC and the LAME codec.
Now put both files into foobar and try to tell them apart in a scientifically proven way...You won't be able to.
Most of my music listening (I'm 40+) tends to be in the car (most cars have lousy acoustics), on the way to / from the pub (on cheap headphones) , or background whilst working on something else or as I'm cooking.
Sometimes I'll sit and listen to music alone, but not often, and so in almost every case conveniece comes above quality for me.
Are my MP3's lossy? Are they compressed? Certainly.
Do I notice in the enviroments I listen to the music in? Sometimes but not very often.
I don't want big files that fill up the phone, or memory stick I have in the car, I want quick access, smalll pocketable devices, quality thats good enought for my use and lots of choice.
I'm glad someone said it.
Each to their own, I hate DIY and football, doesn't mean I don't respect the practioners of said arts. Same with the audiophiles, they get a kick out of it but to most people MP3 is good enough to get to listen to music. Like AC above said, most of us play music in absolutely awful sound environments, like cars. noisy trains and buses or simply on in the background as a pick-me-up to keep you going through a tedious job.
I do film and digital photography and I will happily bore the pants of people about the technical details of that, but most people are happy to snap away to get a general idea of what they want. Does it bother me? Sometimes, yes. I know that with just a few tips their pictures would be ten times better, do I tell them? Of course not ( well maybe when I'm little worse for wear down the local on a Friday night! ). I know they're happy-snappers and if they're happy good luck to them!
For me personally I'm sorry but after 30 odd years of listening to heavy metal an MP3 encoded at 192kbps is good enough. People beating the crap out of their instruments while a man/women screams, yelps and grunts over the ensuing racket, believe me an MP3 is good enough to allow the music to help keep me sane and happy!
So downvote as you wish but please before you do, bear in mind that we all have different priorities in life and for most us we simply need a background tune to make us feel a little better and if a simple, if inadequate, 5MB MP3 does that, where's the harm?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
