But you're comparing different masterings. You also dont know if the CD transport to DAC interface is as clean as the USB input. You'd need to rip the CDs and play each version through the USB input for a proper comparison.
Great video, but whether you call them stair steps or lollipops more bits is going to result in a better waveform. At what point humans can no longer hear the difference between the quality of these waveforms is not explained by noise, frequency, and dynamic range measurements.
The argument is 16 captures it all. There is no finer resolution when it comes to the analog output. The analog output of 24 is the same as the analog output of 16, hence no difference. The same applies to the 44.1 sample rate being able to capture everything and the analog output being exactly the same when you move up from that point. There are two different things being discussed in this thread - mastering/mixing and the final recording. I think the OP was referring to the final recording. Discussing mastering and mixing just confuses things.
I have to admit that the only thing I really know about all this is what I've read and what I've observed from my own recording efforts. I have deduced that a higher bit rates not only improve dynamic range by lowering the noise floor but create a smoother initial waveform at 44/24 because of more accurate calculation of the sampling voltage. The big difference a higher bit rate would make is at the higher sampling rates of 96/24 and beyond. The higher bit rates can more accurately represent the amplitudes of the samples. This is my take on all of this but, in situations like this I always quote Duke Ellington "If it sounds good, it is good"
Yes, true, at the end of the day as "subjects" there will always be some amount of biases in our assessments and I know in the pro-audio and academic world, proper listening tests with dithering algorithms have been done with blinded subjects for decades. My experience has been that dithering of the lowest 1-2 bits of a 16-bit signal is such a minuscule effect and it matters so little audibly that whether someone likes MBIT+ or POWr or just plain TDPF doesn't matter as hobbyists. It's not like the (potential) snake oil we fight over typically. Over the years, we don't get into contentious "dither wars" as audiophiles as far as I recall. In any event, we can easily assess the objective effects of dithering and check whether it's adequate to clean out artifacts and maintain a good noise floor. For example, for those who may not have thought about or seen the effect of dithering, here's a look at what dithering does compared to just truncation of a 1kHz 0dBFS signal using iZotope's MBIT+ module: (Link: imgur.com ) Unless there's a problem or one uses high levels of dithering, dithering generally operates down below -120dB to keep the noise floor clean from truncation anomalies. Hard to get too excited about this as music enthusiasts whether blinded or not! Notice that using MBIT+, the "lightest" noise shaping setting allows us to effectively lower the noise floor in the frequencies where human hearing is most sensitive and shifts some of the noise up >15kHz where we're less sensitive. Regardless, we're looking at a noise floor down into -130dB using this 65k-point FFT. On a side note, SACD (DSD) uses this noise shaping technique to a much stronger degree to push the 1-bit quantization noise up into the ultrasonic region so that within the audible range, much of the noise has been suppressed to better-than-16-bits to qualify as a "high resolution" format.
This was interesting, but was beyond my ability to comprehend since I am not technically oriented. Was he saying that when you take 24/96 file and downsample it to 16/44.1, then invert one of them, and play them simultaneously, you get a perfect null cancellation, which means they are identical?
Direct yes or no question: have you done your own testing? "Big" or "huge" is subjective. What may be insignificant to you is major to others. Maybe you aren't as focused on minutiae as some of us are.
Dither doesn't turn distortion into noise. Dither is noise, and is used to basically cover up the noise. Noise shaping simply moves around the dither so it is less obvious to the ear, or where it will do the least damage to the original sound.
Right. I should have said it replaces the distortion. It's not really covering up anything either. It just randomizes the error, thus avoiding the harmonic distortion and dropouts.
The waveforms will be identical though. That is not how it works. Higher "resolution" is not really more resolution. It simply allows a higher cutoff frequency and a lower decibel limit that is far beyond the noise level of any equipment, mics, mic preamps, Open reel tapes and even any very quiet room. There is no difference in quality of the waveforms. High res does not work the way some think it works, It does not allow MORE resolution with what passes, but simply ups the upper frequency cut off and the extremely low level sounds that are buried in noise to begin with.
Thanks @Archimago for the graph. Been looking around for something like that for awhile as an audiophile chatting on forums and trying to make heads or tails over what people say. To borrow the phrase from someone above, I'm glad that forums these days have more "because...science" people who bother to look deeper into how these things work and explain them well to those of us not as technically inclined. Much better than the decades of exposure to "because...Stereophile says so" comments or heaven forbid "because...Voodoo" people selling crazy sh*i like this (one time I was even interested and fascinated in this stuff).
One should note that such spiky results and high quantization noise power are only obtained on pure signals that are a multiple of the sample rate (limiting the number of discrete values). A 1000kHz tone at a sample rate of 48000kHz only has 48 sample points before the exact values repeat over and over (if not 24). 1001.2939453125Hz tone (an FFT bin center) at 44.1kHz, -0.1dB, using a better window function and proper audio selection with margin outside of FFT window for analysis, and better downsampler: blue: original 32 bit generated tone (note analysis background "noise" would not even appear on your dB scale); green: 16 bit undithered with no DC offset (pure rounding); yellow: 16 bit shaped-noise triangular pdf 0.95bit dither
So how does it not give more resolution? Surely if there is very small alterations in waveforms that are not picked up with 16 bit but could be picked up with 14 bit then the 24 bit will give more resolution? Not saying we can hear that resolution but surely it will be there?
I've heard a difference between the 16/44.1 CD release and the 24/44.1 download of the same album and same mastering a few times. Often enough that I generally prefer getting the 24/44.1 download if the prices for the CD and download are close enough. If the 24/44.1 is priced too high or if it's a release that I want the physical CD and all the liner notes then I'll buy the CD. I assume the difference I'm hearing is due to the dither. And possibly partly due to how my DAC handles 16-bit and 24-bit. The DAC I use primarily is a Schiit Gungnir Multibit which has 19 effective bits of resolution.
It will be there, for any DAC that has more than 16 bits of effective resolution. You can look up measurements for any DAC you are interested in to check the effective resolution.
Really? Under that logic all waveforms would be equal no matter how many bits, so why not use 8 bits instead of 16?
