[Greenears]

 

 

So still what am I hearing?
My best guess lies in a misconception that many have about quantization noise. Please open any standard EE textbook on signals and systems. The first thing you will read is that Quantization errors are a non-linear process and cannot be completely analyzed mathematically. The ~6db per bit idea (which is where you get your 100 dB and 60 dB) is an approximation. I'm not making this up it says it right there in the textbook. I saw a U of Waterloo paper (google) that had a good intro summarizing Q noise, google it. The analogy is similar to FM and AM radio. In EE you learn how to completely analyze AM using Fourier and Laplace. Every detail can be described by nice equations with precise answers. Then the next thing you learn is that FM is a non-linear process that has no equivalent equation. But it sounds better. This revelation is very frustrating to young padawans, but you get over it after a few weeks. A few tricks and approximations and maybe computer simulations are used to analyze FM to enough extent to be able to use it. Same with Q noise - and actually it has some similarities to FM with repeated short spikes throughout the spectrum.

FFT:
You also can't say conclusively you looked at the FFT and didn't see anything. While I agree it's true that you aren't going to miss some 50 dB spike, there are limitations with FFT. Signals move in time, FFT is a slice in time. To convert between domains you need a window like Hann or Blackman and the windows have artifacts. I think everyone that has worked with this stuff hands on knows this.

Possible Theory:
So remember that a frequency shift and phase shift are the same thing (while they're shifting). My suspicion is that human hearing is incredibly attuned to minute frequency differences, which make up what we call "tone". Note how well we pick out peoples voices, or a Stradavarius, or a Gibson. I'm sure you can pick out Mick Jagger or Bono or Bruce Springsteen in the first syllable. No tones are pure, they all have distortion and we can pick out the slight differences in the higher order harmonics. It may be that at some resonant frequencies the quantization introduces just enough frequency shift that you can detect it, or mucks with the relative amplitude of certain harmonics.  

 


I think that you are spot on here in your analysis of why some people can hear a difference. It is about the range of frequencies and how they change. Harmonics are of most importance. What you haven't quite realised is that the information about frequency shift is carried in the shape of the side of a SIN wave, so when you think about all of the SIN waves added together in every sample you will get more 'information' about the sides of waves if you have a more accurate reading and if you have more samples. 24 bit is actually only giving us 20.5 bits whereas 16 bit does give us 15 or 15.5. Assuming that the ADC used for recording was able to provided more than 21 bits of resolution then that is around 25% better resolution. So not massive and CDs sound a lot better through a good DAC anyway. So, it is reasonable to conclude that we are fairly close to getting as good as we can with this technology.

 

Haven't heard the term "side of a sine wave" before.  You mean the first derivative?  Where do you get 15.5 / 20.5 bits from?  if you are saying that quantization errors near the zero crossing cause subtle phase shifts I'm listening but the counter theory is that as long as the INL of the codec is good those errors get filtered at the output and the fundamental comes out unchanged with the errors pushed to higher frequencies at inaudible levels.  But tell me more....

 

[Greenears]

@stv014
 
I hear you on all the amplitude analysis and 0.5 LSB.  But are you sure there are no subtle frequency shifts? The reason I am hung up on frequency/phase is that I am pretty convinced I can't hear 0.5 LSB anything at 16 bit. Most music is compressed to 20 dB DR and I don't know about the pieces I used but even if they are an extreme 50 dB it's still 40 dB above anything.  If you scroll back many many pages people assumed I was hearing noise and again to confirm I am hearing none whatsoever on any clip at any volume. It's got to be phase or I don't know what.

 

[RRod]

  @stv014
 
I hear you on all the amplitude analysis and 0.5 LSB.  But are you sure there are no subtle frequency shifts? The reason I am hung up on frequency/phase is that I am pretty convinced I can't hear 0.5 LSB anything at 16 bit. Most music is compressed to 20 dB DR and I don't know about the pieces I used but even if they are an extreme 50 dB it's still 40 dB above anything.  If you scroll back many many pages people assumed I was hearing noise and again to confirm I am hearing none whatsoever on any clip at any volume. It's got to be phase or I don't know what.

 
You can gradually pin down the samples that are causing the phenomenon by trimming around where you hear the difference. If you trim the difference file at the same time, then you'll be able to see exactly what has been "taken away" from the 24bit file. I would happily post the files I've got, but I would think a complete difference file might be against the rules.

 

[castleofargh]

  @stv014
 
I hear you on all the amplitude analysis and 0.5 LSB.  But are you sure there are no subtle frequency shifts? The reason I am hung up on frequency/phase is that I am pretty convinced I can't hear 0.5 LSB anything at 16 bit. Most music is compressed to 20 dB DR and I don't know about the pieces I used but even if they are an extreme 50 dB it's still 40 dB above anything.  If you scroll back many many pages people assumed I was hearing noise and again to confirm I am hearing none whatsoever on any clip at any volume. It's got to be phase or I don't know what.

I fail to understand how you can thumb up stv014's post(well I do agree with all that he said) and then post this?
how loud do you expect some phase shift distortions caused by a  24 to 16bit conversion to stand?
 
 
 you didn't try to see if there were statistical changes(assuming that your posted abx weren't only your best runs) with and without dither to try and really rule out quantization noise. that's the very first thing I would have done if I happened to get more than guessing. instead now you're hunting for phase.

 

[sonitus mirus]

  @stv014
 
I hear you on all the amplitude analysis and 0.5 LSB.  But are you sure there are no subtle frequency shifts? The reason I am hung up on frequency/phase is that I am pretty convinced I can't hear 0.5 LSB anything at 16 bit. Most music is compressed to 20 dB DR and I don't know about the pieces I used but even if they are an extreme 50 dB it's still 40 dB above anything.  If you scroll back many many pages people assumed I was hearing noise and again to confirm I am hearing none whatsoever on any clip at any volume. It's got to be phase or I don't know what.

 
I'm not sure that people were assuming you heard noise or anything different in your ABX test logs.    

 

[RRod]

   
I'm not sure that people were assuming you heard noise or anything different in your ABX test logs.    

 
More what I was trying to say to him was that all he *should* hear is noise, so that if he was hearing anything else then settings needed to be looked at.

 

[stv014]

I hear you on all the amplitude analysis and 0.5 LSB.  But are you sure there are no subtle frequency shifts? The reason I am hung up on frequency/phase is that I am pretty convinced I can't hear 0.5 LSB anything at 16 bit. Most music is compressed to 20 dB DR and I don't know about the pieces I used but even if they are an extreme 50 dB it's still 40 dB above anything.  If you scroll back many many pages people assumed I was hearing noise and again to confirm I am hearing none whatsoever on any clip at any volume. It's got to be phase or I don't know what.

 
Undithered quantization adds non-linear distortion, which can be subjectively perceived as different effects depending on what the input is. If the input is complex enough and has a high entropy, then the error can still end up being noise even without dithering. But with a low level and/or pure tonal signal, the distortion products (which are similar to those of crossover distortion in that they can be high order and not decrease with lower input level, so it is subjectively a "bad" type of distortion) can become audible at loud enough listening volume.
 
Even if the music has overall not very high dynamic range, the dynamic range can be higher in individual critical bands with instruments that have a sufficiently "pure" spectrum with a low noise floor. So, if there is originally not much content around 3-4 kHz (the most sensitive band of hearing), for example, and the quantization adds a narrow peak there, it might not be masked, and it becomes audible. If you posted some short samples of the test tracks from the parts where there is apparently an audible difference, it could be analyzed better, and other people could try ABX testing the samples, too.
 
As already suggested by others, it would also be interesting to see what happens if you use dither, which ensures that the quantization error is really just uncorrelated noise, and even though its overall RMS level is higher, it is always spread evenly over the spectrum without any narrow band distortion products.

 

[limpidglitch]

   
Undithered quantization adds non-linear distortion, which can be subjectively perceived as different effects depending on what the input is. If the input is complex enough and has a high entropy, then the error can still end up being noise even without dithering. But with a low level and/or pure tonal signal, the distortion products (which are similar to those of crossover distortion in that they can be high order and not decrease with lower input level, so it is subjectively a "bad" type of distortion) can become audible at loud enough listening volume.
 

 
That effect could be heard (and seen) in the examples Rod posted a page back.
A distinct unpleasant sawtooth like sound near the end of the 't' track, audible after 40dB or so of amplification.

 

[RRod]

   
That effect could be heard (and seen) in the examples Rod posted a page back.
A distinct unpleasant sawtooth like sound near the end of the 't' track, audible after 40dB or so of amplification.

 
Woo someone looked at them 

 

[Greenears]

Thanks all for the interesting posts. It seems that some sort of consensus emerges that there might be some audible artifacts on a specific narrow section on undithered if it is unmasked by that passage. When I can I will try abx dithered against both 16 and 24 undithered and see where it leads.

Sorry about the phase sidetrack but I was absolutely convinced by many posters I could not hear 0.5 lsb amplitude. But now I'm reunconvinced. Sic.

 

[bigshot]

It's pretty safe to assume that if you hear a difference between 16 and 24 using music at normal listenng volumes, something is wrong somewhere with either your files or your equipment.

 

[RRod]

Thanks all for the interesting posts. It seems that some sort of consensus emerges that there might be some audible artifacts on a specific narrow section on undithered if it is unmasked by that passage. When I can I will try abx dithered against both 16 and 24 undithered and see where it leads.

Sorry about the phase sidetrack but I was absolutely convinced by many posters I could not hear 0.5 lsb amplitude. But now I'm reunconvinced. Sic.

 
You can hear any signal that is actually there if you can crank the volume loud enough and if the signal is of a type that can stand out above the noise.
 
Here's a 10s, 10 octave sine sweep from 20Hz at -101dBFS in 24 bit.
Here's the same file at 16bit with triangular dither.
 
If you look at the spectrograms you will see the sweep clearly on both, but the 16bit file has noise much closer in loudness to the sweep. So of course if you try to actually hear the sweep, you'll detect noise in the 16bit file you don't hear in the 24bit file. I could do the same thing with 32 vs 24bit as long as I'm making theoretical files on the computer. In either case, at -101dB, you'd have to be listening in an environment at really low SPL to be able to hear anything at full scale without pain.

 

[Stillhart]

Sorry if this has been asked already:  I can't dig through 2000+ posts to find the info.  Is HDTracks.com basically a scam then?  They're selling remastered tracks in 24 bit and that remastering is the difference we're hearing?
 
Also, how does this theory relate to DSD?

 

[castleofargh]

  Sorry if this has been asked already:  I can't dig through 2000+ posts to find the info.  Is HDTracks.com basically a scam then?  They're selling remastered tracks in 24 bit and that remastering is the difference we're hearing?
 
Also, how does this theory relate to DSD?

you should think you're paying for the remastering. and in some cases it's well worth paying as they can sound much better than the CD version. but it's a case by case situation. some remarstering are absolute crap. also some hires albums are just the CD version upsampled so nothing gained there.
that's why it's hard to give a straight definitive answer.
 
DSD is worst, as almost all DSD albums come from PCM masters. so if they touched nothing, you're actually losing "quality" in the conversion(not to mention that a lot of DSD players will convert the DSD back to PCM, so useless double conversion). but just like HDtracks and 24bit albums, some masters will only be available in DSD. so if you want that particular master, you need to get the DSD and a DSD player.
 
but yeah for us, the audio consumer, I would avoid thinking that high res equals to better sound.

 

[Stillhart]

Thanks, that's helpful.  This thread has been eye-opening.  As someone who doesn't particularly believe in the power of $300 cables and pointy things on the feet of my equipment, I'm happy to read scientific proof that some of the questionable stuff is effectively bunk.