'24/192 Music Downloads ...and why they make no sense'

[Thropplenoggin]

http://xiph.org/~xiphmont/demo/neil-young.html

I wonder what the board's resident audiophiles make of this rather intense dissection of the physics of listening vis-a-vis 'studio masters', which are set to become the latest marketing tool of the music industry. I say this because: 1) Sony has a new micro-system capable of playing them coming out; 2) the FiiO X3 is an afforable digital audio player capable of playing them now out; 3) Neil Young's new Pono system is due out in early 2014.

According to this very long and detailed article, 16-bit FLACs are sufficient (i.e. CD quality), so 'don't believe the hype'.

[MrGongGong]

I don't really do downloads or the whole "audiophile" thing
but I can easily hear the difference between 16 and 24 bit audio
most of the time these days I would always edit in 24 bit even if it was going to finally be 16 bit 44.1 (CD quality)
particularly if there were lots of fades to silence etc
all depends on the nature of the signal though and WHAT music you are listening to IMV

"CD quality" isn't automatically very "high"

[Dave2002]

I've seen that article before. There are some good points made, but I do believe MrGG about the use of more bits. I am less sure about using higher sampling rates. There could be engineering reasons why 96ksamples/sec or even 192ksamples/sec might sound better - and this is not to disregard the sampling theorem. Sometimes different kit (for both recording and playback) is simply made better, and that can make a difference - whether based on theory or not. Similar comments apply to recordings - stick the microphones in the wrong places and get the balance wrong, and it won't make much difference what sample rates or bit depth are used.

Not all recordings are made using PCM. The whole business of noise shaping may also make a difference, and that could benefit from high sampling rates.

Mention of new systems (msg 1) is interesting. Although some 16 bit CD quality recordings do sound good, for whatever reasons, some SACDs and other "hi-res" recordings do sound better.

[amateur51]

I don't really do downloads or the whole "audiophile" thing
but I can easily hear the difference between 16 and 24 bit audio
most of the time these days I would always edit in 24 bit even if it was going to finally be 16 bit 44.1 (CD quality)
particularly if there were lots of fades to silence etc
all depends on the nature of the signal though and WHAT music you are listening to IMV

"CD quality" isn't automatically very "high"

But then there is a huge spectrum of human hearing sensitivity. I'm so grateful to have relatively duff hearing and a nature that is grateful for the miracle of recording/performance

[Ferretfancy]

But then there is a huge spectrum of human hearing sensitivity. I'm so grateful to have relatively duff hearing and a nature that is grateful for the miracle of recording/performance

As an ex sound mixer, ams, I had to train myself to stop mentally checking the sound quality all the time when listening to music at home. I still sometimes find myself tempted to dismiss recorded performances if the sound is poor, but generally only with new discs. Historic re-masterings I can accept with no problem.
I tend to be suspicious of claims that this or that equipment is the tops, simply because I'm well aware of the fact that the single biggest factor affecting sound quality is the acoustic of the room, and most people can't do much about that. In any case, large increases in expenditure usually give tiny improvements in what you hear.
Incidentally, I think that headphone listening is distracting, I rarely use mine except on my computer, and then not for quality music. Orchestras are not meant to go right through your head.

[amateur51]

Incidentally, I think that headphone listening is distracting, I rarely use mine except on my computer, and then not for quality music. Orchestras are not meant to go right through your head.

Interesting points from your considerable experience, Ferret

I'm thinking of getting some good quality headphones because my tinnitus may be getting louder. I wasn't thinking of using them for big orchestral pieces but more for instrumental and chamber music.

Apologies to all for this topic diversion

[Dave2002]

throps

Thanks for mentioning xiph.org. This is a really nice video - http://xiph.org/video/vid2.shtml

The amusing thing about that is that the background noise in that (where does it come from?) is far greater than the quantisation or dithering noise discussed in part of the talk.

I really liked the HP signal generator - maybe I should see if they are still available. I bought one for a lab - around 1980ish - looked similar - lovely machines, with an IEEE-488 interface, and we also had some cheap analogue osciloscopes.

It's many years since I did the maths re the sampling theorem, but I do recall that the filter required to get an exact rendition of the input signal was a particular one - just any old filter won't do. In practice though, the errors are likely to be small if a "good enough" filter is used.

Despite all that though, I still feel that there may be benefits to so-called hi-res masters and equipment. There is undoubtedly some hype, and some marketing influence, but even allowing for that I believe that there are benefits. It's very hard to prove though, as "golden ears" may be needed.

Even though I don't have such ears, I can still tell the difference between different audio systems, and generally between live and reproduced sounds.

[Dave2002]

I'm thinking of getting some good quality headphones because my tinnitus may be getting louder. I wasn't thinking of using them for big orchestral pieces but more for instrumental and chamber music.

Note that ferret does not particularly recommend headphones. I use them myself, and find them useful, but like Tony Hancock in The Radio Ham, after a while one's ears get hot. I think they can also aggravate any noises you may experience within your ears/head if used for long periods, so take care.

I do recommend them, and the sound quality can be high compared with loudspeakers, but one does need to take care. Also, volume levels can be really problematic, particularly if computers are involved, as it's possible to inadvertently get very high volume levels (for example if a computer alarm comes on - thanks Apple/Microsoft - do we really need a "boing" when mail comes in?) which are (a) painful, and (b) maybe very damaging.

[Gordon]

It's many years since I did the maths re the sampling theorem, but I do recall that the filter required to get an exact rendition of the input signal was a particular one - just any old filter won't do. In practice though, the errors are likely to be small if a "good enough" filter is used.

Yes, Dave the maths does place conditions on the filters and the original Nyquist/Shannon work assumed perfect square box shapes at both ends so the impulse response of the channel is well defined as a sinx/x shaping. Mathematically this provides an exact reconstruction - ideally the filters have linear phase, another Nyquist condition. One of the beauties of linear systems is that one can correct upstream errors downstream but for that one needs full knowledge of that upstream history.

Another mathematical feature of the sampling theorem is that if the passband [ie the complete channel again including both ends] of the filters used is symmetric about the half sampling frequency then correct construction is also possible. IOW one does not need square box rapid roll off. This is not invoked for audio. See here at page 20 or thereabouts:

http://www.amazon.com/Digital-Transmission-Engineering-Series-Communication/dp/0471694649

The theory also assumes perfect practical implementation which clearly may not be possible. Apart from the sampling frequency value its stability matters a lot for audio and so any jitter [which is not mentioned in the classical theory] will add its footprint to the reconstructed audio. This is not a matter of theory but of engineering. We've discussed this before and some while ago I posted [can't remember where the thread is] some figures illustrating the degree of jitter suppression needed. It is surprising how small jitter is before it becomes significant.

In practice the anti-alias and reconstruction filters, one at each end, are supplied by different designs and are not necessarily complementary and so the channel shape isn't defined to the required mathematical degree. Whilst they might approach the ideal shape, as you imply they may be "close enough" to provide a given amount of precision. All this applies to any sampling process but of course any distortion in between the filters will also add its effect even if the filters are perfect. So whilst we apply the theorem to PCM [ie 44.1/16 CD] it also applies to DSD but that is a much more complex situation with sampling frequences much higher than Nyquist.

Incidentally, as MrGG implies above, the spectral properties of the signal itself are relevant. If you apply the sampling to analogue composite video [eg the old analogue PAL system now obsolete since 2012] theory suggests a need for a sampling frequency of at least twice 5.5 MHz, the bandwidth of UK PAL. However, the presence of the colour sub carrier at 4.43361875 MHz allowed a bending of the rules such that sampling at twice this frequency, rather than the maximum frequency present in the video, was sufficient. THis was at first sight sub-Nyquist but it worked - see any number of BBC Research papers from the late 1970s or download Volume 12 from here:

http://www.ntlpa.org.uk/memorabilia

The process of sampling and band limitation of a signal and then its quantisation are two totally separate processes not to be confused, at least for PCM . DSD does do that to some extent. MrGG is right that when mastering audio from multiple sources 24 bits and a high sample rate would be best but then, once finished, the audio at CD level will be adequate for most people's ears but not necessarily for audiophiles who will require well engineered hardware and good acoustic listening conditions.

[MrGongGong]

particularly if computers are involved, as it's possible to inadvertently get very high volume levels (for example if a computer alarm comes on - thanks Apple/Microsoft - do we really need a "boing" when mail comes in?) which are (a) painful, and (b) maybe very damaging.

The first thing to do when getting any computer
is to turn ALL of these off

(I also turn off all the 'sleep', 'hibernate' etc things but that has more to do with using the machine for gigs and having the scary experience of going onto the stage at the RAH to find everything had 'gone to sleep' ....)

Also there are headphones and there are headphones

For detailed listening I have a pair of these



Light, designed for mastering etc so flat response and less tiring than most , you would need serious money (and understanding neighbours) to get speakers at that quality.

[Dave2002]

One of the beauties of linear systems is that one can correct upstream errors downstream but for that one needs full knowledge of that upstream history.

Indeed. I once had a discussion with some simulation modellers at NASA. They had noted that they got exactly the same numerical results with different models, but as they had non engineering and non mathematical background this was a curiosity. I followed it up, and one of my friends pointed out that since they were modelling by using linear models it made no difference in which order the different parts of the system were applied. This of course explained the exact nature of the results.

I am just mulling over the other part of your sentence. Is it not the case that if there is a zero in one of the functions that complete correction of upstream effects is not possible? Functions without zeros should be correctable, though ill conditioning may make this difficult in practice. Maybe we are using slightly different terminology - I could be expressing the concept incorrectly.

[Gordon]

I am just mulling over the other part of your sentence. Is it not the case that if there is a zero in one of the functions that complete correction of upstream effects is not possible? Functions without zeros should be correctable, though ill conditioning may make this difficult in practice. Maybe we are using slightly different terminology - I could be expressing the concept incorrectly.

Maybe we are at cross purposes!! What do you mean by "functions without zeros"? Stability theory of systems uses mathematical analysis based on the real/complex plane and in that case where the poles and zeros of the complex response are in that plane matter. Is that what you mean?

I'd have to go back to the Nyquist theory to see what actual conditions are placed on the passband response shaping. By and large the filters used for ADC and DACs are well behaved - eg Butterworth that has a flat pass band - but to get the fastest roll off they often have passband ripples whose amplitude must be small which suggests lots of poles/zeros. Try looking up Tchebyshev or Elliptic Function filters on Wiki - there may be some stuff about the pole/zero business.

[Ferretfancy]

Note that ferret does not particularly recommend headphones. I use them myself, and find them useful, but like Tony Hancock in The Radio Ham, after a while one's ears get hot. I think they can also aggravate any noises you may experience within your ears/head if used for long periods, so take care.

I do recommend them, and the sound quality can be high compared with loudspeakers, but one does need to take care. Also, volume levels can be really problematic, particularly if computers are involved, as it's possible to inadvertently get very high volume levels (for example if a computer alarm comes on - thanks Apple/Microsoft - do we really need a "boing" when mail comes in?) which are (a) painful, and (b) maybe very damaging.

Good points. I think that my main reservation about headphone listening is that most recordings are monitored on speakers and are designed to be heard that way. I know that headphone quality can be very good, and that modifications are available to create a soundstage in front of the listener, but I think the argument still stands.
As a point of interest, I have just bought the last CD in the Bach Cantata series directed by Suzuki, and these BIS discs are in compatible SACD, monitored on STAX headphones ( perhaps not exclusively ) Heard in stereo they are excellent performances, but the reverberation does rather overwhelm the soloists, and I can't help wondering whether the headphones are a factor. Of course, it may simply be that heard in SACD surround the problem does not arise, I have no way of checking unfortunately.

[richardfinegold]

I can easily tell the difference between most downloads in 24/192 or other high resolution formats and CDs. i say "most" because some high res downloads are dissapointing (I just downloaded Cedille Records Pacifica Qt of DSCH Qts and find them no better than several versions that I have on CD in terms of recording quality), and some CDs sound exceptional. Bit rate doesn't always tell the whole story.
What I can't tell the difference is between high Res downloads and the same recordings issued on SACD. I therefore will continue to invest in SACDs and hope that they remain available, because as I have written before, I intensely dislike downloads.

[MrGongGong]

Pauline Oliveros talks about being banned from a studio when she was a student because she was including ultrasonic sound in her pieces.
Whilst they ARE (by definition !) out of range given the nature of sound surely they will have effects on the sounds in the audible spectra ?