Open both in Audacity, invert the polarity of of one, and sum the two into a new track and look for any differences.
Is there some bit rate where the Audacity inverted polarity test does not sum the difference to null? 14 bit? 12 bit? 8 bit? 4bit? I would think there would be audible differences, but if the audacity test still sums to null, I’d question its validity, and vice versa.
I'm not sure what you are asking. Audacity doesn't sum to null unless they are exactly the same. If there is a difference between the 24bit generated waveform and the 16bit you will see it. Try it yourself. It's not hard to do at all. I did it with the test files posted earlier.
I am asking whether “summing to null” is a valid test of “sounds the same.” Many people claim to hear a difference between 24 and 16. I thought a way to test that is to compare 16 bit with 8 bit, which I would think are audibly different. If the test summed to null, that would seem to invalidate it as a test of “sounds the same.” If it did not sum to null, that would tend to validate it. The reason I don’t try it myself is that I am not versed on Audacity.
I did a quick comparison in Audacity: 1) file A - file B (difference between 16 and 24 bit): RMS of the resulting difference (aka quantization noise) = -96 dB 2) file B - file B converted to 8 bits (difference between 16 or 24 bits, and 8 bits): RMS = -47 dB 3) file A - file A converted to 8 bits (difference between 16 or 24 bits, and 8 bits): RMS = -47 dB I did the conversion to 8 bits both from A and from B, because I didn't know which one was the 16 bit and which one the 24 bit. Turns out it doesn't matter (which is further proof of the indistinguishability between 16 and 24 bit). A -96 dB noise floor is inaudible by ANYONE. -47 dB, on the other hand, is definitely audible. For comparison, a turntable with a MM phono stage has a noise floor around -60/70 dB (but in practice often worse due to surface noise)... and we all know how much vinyl noise is noticeable. In other words, we are saying that to have at least a slight chance of hearing the difference in quantization noise between 16 and 24 bit, you would have to crank the amp so much that the music would be at a level that's so loud as to be not only uncomfortable, but actually DANGEROUS (as in: permanent damage to your ears in half an hour). On the contrary, the noise from a quantization to 8 bits would be at the level of the noise in a normal urban home when listening to music at a normal loudness. Curiously, the theoretical best noise floor for 8 bits would be -48 dB, so the real world result of -47 (actually -46.9) makes a lot of sense.
So in other words it didn’t sum to null when inverted polarity? Forgive my ignorance, but are these all 44.1 kHz?
All at 44.1 kHz, yes: it has no impact on quantization noise. And yes, it doesn't sum to "null". It sums to -96 dB in one case (which is equivalent to "null" in real world, see post above about how inaudible that is), and to -47 dB in the other case (and that, instead, is as audible as the car traffic on the street when staying indoors with the windows closed).
Play the video from earlier in this thread: 24 bit vs 16 bit music files It should start at the point when he compares the two files in Audacity. If you want you can start the video from the beginning. Edit: to be honest that video isn't as simple as it could be. He should have merged the two into a new track. You'd then see a flat line, meaning no audible difference. It's super simple to do in Audacity. You open both files (24 & 16), invert one, and merge the two into a new track. That's it. Audacity is free. Download it and experiment with it. Open a hi-res file, then downsample it to 16 44 and save it. Then open the original and the 16 bit, inverse one, merge and see for yourself. You won't see a difference. The reason is that although the hi-res and 16 bit source are different, the resulting audio isn't. Why? Because 16 bit 44 captures all of the audio. That's why CDs are 16 44. They would have gone higher if it made a difference. It's difficult to wrap your head around it, but audio is different than video/imaging. There are no pixels at the analog output stage. What you see in Audacity is the analog output, so when you sum them, you will see any differences, and between 16 44 and hi-res you won't see any.
If you take two identical signals, invert one and sum them they'll cancel each other out. If you take two identical audio files, invert one and sum them you'll get silence (to the precision of the math, probably around -300dB). If you take 24bit and resample it to 16bit you will not get a full null, but you will get a null down to where the differences between the signal is, which is the 16bit noise floor circa -96dB. 24bit vs. 8bit will null to the 8bit noise floor circa -48dB. And yes, it's a perfectly valid test. Taken a step further I posted a file earlier in this thread which is the null of 24/88.1 and 16/88.1 played through a DAC and recaptured at 24/196. That's a worst-case scenario as the signal is fully reconstructed and re-sampled and yet the difference is some -96dB.
Exactly, as proven by this further test in Audacity: 4) file A - file A: RMS = -inf Again, in other words, MATHEMATICALLY the difference between 16 and 24 bit in Audacity is -96 dB, and not -infinity. So, mathematically, there IS a difference between 16 and 24 bits. But the difference is so small that you would have to crank your amp to something like +120 dB RMS to have a chance to hear it. But... 1) the hiss from the amp would probably STILL be louder than the quantization noise and completely mask it, with that level of amplification 2) your ears would probably start to bleed (and not figuratively) much earlier than that...
Enjoy the difference file (16 bit - 24 bit) of the Daft Punk track, exported from Audacity: difference.wav Crank it up as loud as you want. The only thing you will hear is the hiss of your amplifier (but then remember to turn the volume down before listening to anything else!!!!)
If we accept that 24/44.1 can’t be meaningfully distinguished from 16/44.1 on the inverted polarity test on Audacity (meaning, I think, they both sound the same for all practical purposes), does it necessarily follow that 24/96/192 cannot be meaningfully distinguished from 16/44.1 on the same test? I think that’s the more heated debate on this forum.
Yes, 16/44.1 captures everything. Think of it like a bucket that holds 10 gallons of beer. You have a 15 gallon keg of beer with a tap and you fill up your bucket. You also have a water tower with 1,000,000 gallons of the same exact beer, and you fill up a 2nd bucket with it. The end result is two buckets of the same exact beer.
As I said days ago, give me 16 bit/96 kHz any day, over 24 bit / 44.1 kHz. Difference between 16 and 24 bits cannot be heard, by no-one (even those who say otherwise). Difference between 44.1 and 96 kHz can be heard indeed by someone with a trained ear and not too advanced in age, although it's mostly due to the imprecise implementation of most DACs/players rather than to intrinsic limitation of the 44.1 format. A good oversampling DAC will play 44.1 as beautifully as 96.
I was surprised just how good the 8 bit file sounded when I tried this. Obviously the noise floor sounded like a cheap cassette with no noise reduction but the actual music sounded the same as the 24 bit file for louder music!
I don’t recall ever seeing a 16/96 track on Qobuz, which is my hi rez streaming source. Didn’t realize it existed. FWIW, I can say that upsampling a redbook cd on a friend’s DCS DAC to the higher sampling rates is clearly audible. Sometimes “better” to my taste, sometimes not. But definitely different.
Never seen it either on ANY platform, and it's a shame. If you want high res, you have to take the complete package (including the useless extra 8 bits). And sure: resampling does NOT add anything above 22.05 kHz. But at least it makes the phase shift at high frequencies much more tolerable (less ringing and brittle artefacts).
I am also a photographer, and it's very much similar. 10 bits per color channel is PLENTY for displaying ANY photo. And yet, cameras can record at 12 or 14 bits per color. And the difference between 12 and 14 bits can only be seen during photo editing, and only when doing extreme stuff like raising shadows +6 stops (which is equivalent to a 64x amplification). And even then, it is only slightly noticeable as a minor difference in the rendition of noise. But for viewing the final result, 10 bits remains more than enough. So much more than enough, that normally only 8 are used (jpeg...).
That would be my main conclusion from the above, then... redbook/normal CD quality is quite sufficient to capture everything useful, one can always implement oversampling in the playback machine for improved sound - or is this not a good idea for other reasons??
It's a perfect idea, since nobody can hear above 20 kHz anyway! But, quite simply, no CD player from the '80s and '90s could do it... And that's why CDs got bad rep from audiophiles for the "harsh" highs. I still think that a true 96 kHz recording, or a DSD one, can still have a marginal edge though: it sounds slightly more natural than oversampling. But hey, we are speaking about very subtle differences. I own 600 CDs and 1000 LPs, and I have never really bought the opinion that LPs are better with high frequencies because they don't have a hard frequency cut. Sure, LPs have a more gradual and natural frequency falloff compared to the abrupt and distorted one of old CD players. But it's an unfaithful rendition of those frequencies all the same: it may be more pleasant and easier on the ears than 44.1 digital (and I wouldn't have 1000 LPs if it weren't so), but it's NOT faithful.
My first cd player was the original Adcom cd player. It had an “analogue” button that brought in reduced channel separation and adde ambient and slightly out of phase stuff to make it sound more like an LP. I would not deduce from that that the LP is a more accurate source of reproduction.
Thanks for the results so far guys... Listen to the tracks and let me know your impression on the 16-bit vs. 24-bit test. Stand up and be counted! The way I see it, this is your chance to say "I listened and did something to convey my impressions." instead of a person who sat around and either just say things on audiophile forums or be accused of being a "science guy" who never actually listens.
Taking Archimago's files, waveforms null, yet spectrograms clearly different, how can it sound the same?