24bit vs 16bit, the myth exploded! - Page 70

Quote:

Originally Posted by gnarlsagan 


Kiteki made a point in another thread that we have no way to measure a driver's tone. For example a way to measure differentiation between a titanium driver and a carbon nanotube driver. They sound different in some way that isn't measurable.
It was also said that there is no way to measure the difference between a flute and a trumpet in way that would tell us which was which. They have different tone in a way that isn't measurable.
I'm not agreeing with this at all, but I'd like to know what more knowledgeable folks than me have to say about it.

Wrong, they sound different in a perfectly measurable way: different harmonic contents, specifically D2/D3/D4/D5 ratio.

Same with flute vs trumpet. Flute has nearly no higher and odd order harmonics, while trumpet has lots.

Also the flute has more noise, specifically pink-like noise and sharper attack than that attainable using the trumpet without overblow.

I definitely agree with AstralStorm!  But the difference between the two drivers will be from lots of mechanical and construction factors not just the titanium or carbon nanotube material.
 

Quote:

Originally Posted by emmodad 

 

 

[ cue wide-eyed innocent face ]

 

 

oh my goodness.

 

 

this couldn't possibly be interpreted to mean that there might be fact bending, misrepresentation and selective omission "to fit the argument" in the audio hobbyist and audiophile world?

 

heaven forbid such nefarious tactics in places ie perhaps

 

- the a/m xiph manifesto / positioning statement for a certain lossy coding implementation, timed and published by a principal of that technology in the face of pending commercial irrelevance;

 

- the OP's ... ahem,  interesting...  presentation of digital audio and signal processing basics in this thread;

 

- some of Those Audio Specialty magazines and a few "let's leverage the trend and most people's lack of the difficult technical knowledge" computer-audiophile-type websites with "well-known" and "highly-experienced" (and yes - very prolific in their posting and public self-promotion....errr, presence) -  writers and "founders" (holders of any actual technology background education and knowledge:  few; acquirers of clicks, pageviews, advertising revenue and high-end audio gear:  many) who coordinate with a small cabal of similar-intentioned (and often technically suspect) "specialty" equipment manufacturers.

 

 

[ cue image of Opus in mode of holding his chest for pain of disappointment rending his ever-pure heart ]

 

 

and as a public service reminder to all about the technical misinformation which started this entire thread:  if you'd like some reasonably-digestible and technically-correct resources about underlying theory wrt bit depth, resolution, dynamic range, sampling theory and digital signal processing in general, wander on over to check out the links in some posts earlier in this thread @ 463; 562 (incl links to references via thread "Bit depth and sampling frequency explained"; 651

 

 

 

now let's all go lie down in a field of Dandelions...

hey emmodad, I found the sample sections of the books you mention in the other posts very informative, esp. the Pohlmann book. Lots more liberating than reading a thread full of bickering, I might add.

Wow, it has taken me a while, but I have finally read all 70 pages of this thread. My (non-professional) takeaways:

• The only difference between 16 and 24 bit is the dynamic range (distance between loudest and quietest sample). Almost all of this falls below the threshold of audibility, and even within the range of hearing, the difference in gradation given by the extra bits in 24 vs 16 is so small as to be undetectable.
• Sampling at a given rate should, theoretically speaking, allow for the perfect reconstitution of a signal with a frequency of half that rate.
• There may be some evidence that despite the above, some technical benefit is gained for playback with sampling rates greater than the 44.1/48 that should encapsulate the entirety of human auditory response.
• There is a law of diminishing returns with sampling, beyond which any possible benefits are outweighed by the potential introduction of errors and computational requirements.
• There may be a correlation between sampling/bitrate and perceived quality, but this is almost certainly due other factors, such as quality of original recording equipment, mixing, and mastering, and not the actual sampling rates and bit depths.

In a nutshell:

1. Well recorded, mixed, and mastered material (at >= 24/96) if properly downmixed and dithered to 16/44.1 will be functionally indistinguishable from the original recording.
2. Music should, as often as possible, be played back at the sampling rate at which it is recorded to minimize resampling errors
3. There may be reason to believe that commercial music offered at 24/96 may offer a better experience due to the extra care that likely went into its creation, but such music, properly downsampled, should be indistinguishable from the original.
4. Anything greater than 24/96 is overkill on the level of swatting a mosquito with a BLU-109.

On a personal note, I have seen people refer to potential benefits of higher sampling rates capturing sonic characteristics such as "airyness" or "space". I am not sure what causes the brain to perceive these characteristics, but, bottom line, if all we can pick up is compression and rarefaction of a medium in a set band of oscillations, and the digital signal can be reconstituted so that when it is applied to a membrane it causes the exact same compression and rarefaction as the original, shouldn't that encapsulate ALL the information that was encoded?

<null edit>

Sub.

I love your summary, AVI. 

Just my contribution: I've dabbled in upmixing my music myself, and I often like it. You might laugh at me for saying this, but there is a subtle difference, in that it seems like I hear more detail. Determined with ABX testing. 

Let me explain:

In my photography analogies, I prefer to process my photos with Photoshop's sharpening filters. Much of it helps with interpolating, and therefore, reducing effective lens fuzziness. Also, higher contrast/more edgy edges, makes details easier to notice. Is the latter preserving the original picture's information? Nope. We are altering it. But it makes it easier to notice. Good for perception. Sometimes to help our imprecise senses, the original data needs to be "helped."

A linear phase brick wall filter (one that has constant non-zero gain up to half the sample rate, then zero above, and the same phase delay at all frequencies) is actually the mathematically "correct" way of reconstructing the continuous time signal, and turns a digital waveform of 0,1,0,-1,0,1,0,-1,... into a perfect Fs/4 sine wave. In practice, it does not have an infinitely steep roll-off, because that would require using an infinitely long impulse response, and a less steep roll-off also reduces ringing.

I've read this entire thread too 

I tend to let my ears and my speakers and headphones guide me

Right now, I fall firmly into the 24bit supporters camp

Granted, most of my digital library is 16bit ripped from CDs, but my 24bit HD downloads sound superior

That could be due to the masters used to produce those versions, but ... I'm not so sure

FWIW my computer system is pretty revealing: Mac Mini > Stelllo U3 > W4S DAC-2

P.S. the whole 44/48 vs 88/96 KHz is still a wide open discussion for me

Well, having skimmed much of this thread, I'm not going to comment on people's perceptions...

However - for anyone who actually wants to learn about signal processing techniques, there is a great opportunity coming up soon - starting in February is this great free course on Digital Signal Processing: https://www.coursera.org/course/dsp

Coursera is a free online program that partners with universities to offer some of their courses in a free online format, where you learn alongside the "real" students, meaning very high quality courses. You don't get college credit for it, but do get a certificate of completion. The class is 8 weeks, and includes electronic textbooks.

The professors teaching it are well known and respected in the field, and the course is intended to cover signal processing theory (rather than specific implementation) which makes it very applicable to those of us who are interested in the concepts as they apply to DSP chip, CPUs, etc. Highly recommended for anyone who thinks they know a lot about this stuff, or is aware that they don't. ;)

-Sam

Thanks for the heads-up phidauex

No need to read entire thread after the first few pages.

Dan Lavry: http://www.monoandstereo.com/2008/06/interview-with-dan-lavry-of-lavry.html

"Regarding bits: The ear can not hear more then about 126dB of dynamic range under extreme conditions. At around 6dB per bit, that amounts to 21 bits, which is what my AD122 MKIII provides (unweighted).

Regarding sample rate: The ear can not hear over 25-30KHz, therefore 60-70KHz would be ideal. Unfortunately there is no 65KHz standard, but 88.2KHz or even 96KHz is not too far from the optimal rate."

Bob Katz: (at whose feet the OP allegedly worships): http://www.tnt-audio.com/intervis/digidoe.html

"My conclusions are that the wordlength increase is the most dramatic improvement, with the sample rate increase being a secondary factor.
I like the results, but after a number of careful tests, I feel that 44.1 kHz/24 bit can be considerably improved, never as good as 96/24, but a lot better than people think. I am displeased with the losses I have heard with the current generation of sample rate/wordlength converters, when my 96/24 recordings were reduced to 44.1K. These converters are constantly improving, and you have not yet heard these recordings at their best in the CD format." 

The Katz interview is fairly old now, and even though A/D converters have improved dramatically, Katz still prefers 24-bit overall because of its superior resolution. On October, 18th 2012: http://www.digido.com/

In regards to 24-bit, "The improvements with higher sample rates go down exponentially with each higher rate. The most significant improvement comes from moving to 48 kHz from 44.1 kHz. 88.2 and 96 k are not "twice the improvement" of 48 k, but they are a meaningful improvement."

Anonymous internet poster < Bob Katz, Dan Lavry

I wonder how Lavry came to the conclusion that human ear can actually hear anything at 25 kHz.

Scientific consensus is that the extreme is 21 kHz. (actually something right beyond 20 kHz)

This means a nearly perfect reconstruction filter is possible from 44100 Hz, which has a bandwidth of 22 kHz. There's even 4% rolloff to work with...

Similarly, Katz was talking about the reconstruction filters of the time, usually too cheap (analog?) and too rolled off.

Still, why is he holding onto his opinion now that the filters are much more refined in the better hardware?

Or are they advocating the analog filters, which are minimum phase, but have phase shift at the highest frequencies?

These become viable only when 48kHz is used, otherwise there is major phase shift in audible spectrum.

I'd like to see a controlled experiment.

Not only does it not make sense that he said humans can hear up to 25-30KHz, but how on earth does that equate to an up to 70KHz sample rate? 

the funny thing is that people who claim they hear a difference between 24/96 or 16/44 or between Flac, wav or mp3 will probably fail a blind test., I've downloaded HD tracks  of some mp3 song that I have to check if I could hear a difference, and there was no difference even the piano instructor at my college couldn't detect any difference and he knows one or two things about music, and I think if you are paying a lot of money for HD tracks you should hear the diference right off the bat, and this goes to the people who thinks that cables and burn-in make any difference too.