Can you hear the difference between 16bit and 24bit audio files?

[Skeptic]

I've tried similar products in the past, and while they help a bit, they do nothing to alleviate the problem of headphones being able to project an image either in front of the listener or behind the listener.

Well, it depends what you have used, but previous products were usually just "crossfeed" which is a very basic approximation.

Newer products use HRTF processing, and can use other processing to simulate a room/speakers. Especially with a multichannel source, these can be quite effective.

 

But you absolutely do need some kind of DSP for headphone listening, if you want it to sound anything like speakers.

Simply sending them a stereo signal is always going isolate each ear and break how we naturally locate sounds.

How far "outside your head" things can get depends on your equipment.

[gmgraves]

George

In general this is the case, but using decent headphones with a better than average headphone amplifier, some recordings such as "The Storm" from a Chesky Hybrid SACD can sound frighteningly real , with both depth and the illusion of height.

 

Regards

Alex

 

 

I have a pair of HiFiMan HE-500s; they are superb isodynamic phones powered by an excellent HiFiMan EF-5 headphone amp. While they do sound spectacular, they don't image any better than any other headphone which is my general complaint with all headphones regardless of make model, cost or quality.

[gmgraves]

Well, it depends what you have used, but previous products were usually just "crossfeed" which is a very basic approximation.

Newer products use HRTF processing, and can use other processing to simulate a room/speakers. Especially with a multichannel source, these can be quite effective.

 

But you absolutely do need some kind of DSP for headphone listening, if you want it to sound anything like speakers.

Simply sending them a stereo signal is always going isolate each ear and break how we naturally locate sounds.

How far "outside your head" things can get depends on your equipment.

 

 

I suspect so.

[sandyk]

I have a pair of HiFiMan HE-500s; they are superb isodynamic phones powered by an excellent HiFiMan EF-5 headphone amp. While they do sound spectacular, they don't image any better than any other headphone which is my general complaint with all headphones regardless of make model, cost or quality.

 

George

A friend of mine bought a pair of those , but although they were very good in the midrange , they seemed to lack something in the HF area. We both agreed on that. He sent them back for a refund and got a pair of Stax SR-507 to use with with a DIY Kevin Gilmore SS amplifier

 

 

Regards

Alex

 

[bernier]

hi all, just for info sharing i found about 16bit and 24bit

[gmgraves]

George

A friend of mine bought a pair of those , but although they were very good in the midrange , they seemed to lack something in the HF area. We both agreed on that. He sent them back for a refund and got a pair of Stax SR-507 to use with with a DIY Kevin Gilmore SS amplifier

 

 

Regards

Alex

 

 

Must be your taste in HF at work. The HE-500s are reasonably "headphone flat sounding" to about 35 KHz. The silky smooth top -end is one of the things I like best about the HiFiMan headphones. Most headphones have a very ragged high-frequency response which makes them sound unnaturally bright to me. Having said that, The HE-500s seem to be somewhat amp sensitive and don't sound as good when plugged into most headphone amps or headphone jacks on some components. With the HiFiMan EF-5 as well as the Schiit Asgard2 headphone amps the HE-500s really sing, though, so that could be what your friend was experiencing.

 

BTW, that Stax FR graph isn't all that great in either the upper midrange or the highs, is it? While the Stax SR-007s and SR-009s are magnificent phones and for sound quality, I don't believe that anything can touch 'em (for the price they ought to provide sex along along with the great sound though). however, the need for a special, heavy and expensive extension cable in order to be useful is oft-putting for me (I don't mind the power supply requirement for ES headphones, most phones sound their best when powered by a dedicated headphone amp so, it's six of one and half-a-dozen of the other. Again, Stax prices for their amp/power supplies is somewhat high, but I guess they're worth it).

[sandyk]

Hi George

I know which one I would prefer, at least based on the graphs, if I hadn't heard both through a couple of different HAs.. Stax are in general, renowned for how they portray the differences between female songstresses.

The higher priced Stax headphones are of course, even better than this relatively cheap model.

Regards

Alex

 

[Teresa]

...I just read the User's Guide to AudioGate, and it seems to be fine to go up to +3dB‐SACD. It was some of my PCM 24 bit files from non-audiophile labels that went to +1 or +2 dB‐SACD, I didn't notice any of the audiophile recordings going over 0 dB-SACD. However that +1 to +2 dB-SACD alarmed me, so I backed down to -3 dB‐SACD.

 

From the AudioGate user's guide:

 

During song playback and during export, AudioGateʹs level meter shows the peak values of the output data following sample rate conversion...

 

Normally, you will adjust the GAIN (see page 21) so that the peak value of the 1‐bit output data stays within the 0–+3 dB‐SACD range.



It is not necessarily the case that the output signal has clipped at the point that 0 dB‐SACD or +3 dB‐SACD is exceeded; however when the integrator of the Delta‐Sigma modulator reaches a specific amount, the clip indicator will light and the signal will be output in a clipped state.

 

The conversion algorithm of most PCM <‐> DSD converters (including AudioGate) is designed so that by default, 0 dBFS = 0 dB‐SACD.

 

None of the files I converted to DSD128 caused the Clip light to come on, looks like I could have left things at the default 0 dB‐SACD.

 

Update to those interested.

 

I'm using the default 0 dB‐SACD setting and watching the meters as some of the albums I converted at -3 dB‐SACD lost a little bit of their heft and punch, which is there at 0 dB‐SACD. Also at the default 0 dB‐SACD the DSD128 (5.6MHz) file has the same level as the original 24-bit WAV file.

 

I am watching the meters and so far the only files that have went over 0 dB‐SACD are from the major labels and so far the highest is +1 dB‐SACD and Audiogate says DSD is good to +3 dB‐SACD. If any file clips I will redo the whole album it at a lower level.

 

There is some great sound buried in even high resolution PCM files that DSD brings out. I think it might be the elimination of the sharp filters PCM uses. And at DSD128 (5.6MHz) the DSD noise doesn't even start until 60kHz.

[Teresa]

I reactivated my Twitter account so I could try up-converting to DSD with my original Audiogate 2

 

Final Update: The experiment is over, and I once again deactivated my Twitter account as I don't use social media. I only used Twitter in conjunction with AudioGate. While I do like the sound of 24 bit PCM converted to DSD better, it still doesn't have the effortless, natural, musical sound of DSD from well made analog or pure DSD masters. So, as I stated in the past, going forward I will only purchase DSD downloads.

[Fokus]

I think it might be the elimination of the sharp filters PCM uses.

 

But the conversion of PCM to any low-bit high-rate format, including DSD, requires reconstruction filtering at half the original sample rate. The filters are not removed at all.

[Teresa]

But the conversion of PCM to any low-bit high-rate format, including DSD, requires reconstruction filtering at half the original sample rate. The filters are not removed at all.

 

I know the PCM filters are there for recording, but would the PCM filters still be required for playback after conversion to DSD? I thought slow analog filters for DSD would be used instead. Anyway as I stated in the future I'm only purchasing DSD downloads.

[Fokus]

would the PCM filters still be required for playback after conversion to DSD?

 

These filters are a necessary part of the very conversion of PCM to DSD. Whether you do the filtering within a delta-sigma or bitstream DAC chip during playback, or off-line, during conversion of a PCM file to a DSD file, makes no fundamental difference.

[Teresa]

Two weeks after my final update:

 

I went legit

 

I’ve read posts on the internet that clearly state that if one sells or deposes of a recording they are not legally allowed to keep and listen to a copy of it. The more I thought about that, the more I tend to agree with the statement. Legal downloads should be legally purchased, sample promos from the recording company or samples from legitimate websites or recordings one physically has in their possession. My conscience has finally gotten to me.

 

I had extreme financial difficulties four years ago and had to find more and more things to sell. So I sold my Music Hall turntable, Nakimichi cassette deck and modified Yamaha SACD/DVD-Audio player and all my LPs, cassettes, SACDs and DVD-Audios. Before I sold them I ripped my favorites at 24/96 using the analog input and internal ADC on my Mac Mini with Audacity software. Since I’m not legally entitled to listen to these rips anymore, I have been gradually replacing them with legal DSD downloads, mostly from Acoustic Sounds Super HiRez web store.

 

Well, I have bitten the bullet and deleted all of the rips I made of recordings I sold. Now my DSD downloads greatly outnumbered my remaining PCM ones, and going forward I'm only buying DSD downloads.

[Paul R]

Except part of the process involves upsampling first to very high sample rates.

 

As well, part if the current thinking is that during playback of DSD, the filters involved can be rather simple analog filters. In any case, filtering DSD is simpler than filtering PCM, and probably far less detrimental to the sound.

[Fokus]

Except part of the process involves upsampling first to very high sample rates.

 

Except.

 

Except what?

 

Say you have a signal sampled at 44kHz. It defines a spectrum from 0 to 22kHz. From 22kHz to infinite it defines a spectrum comprising of the 0-22kHz baseband positively and negatively mirrored around integer multiples of 44kHz. That is a fundamental mathematical fact.

 

If you regard the signal in its (limited) 44kHz sampled signal space all you are aware of is 0-22kHz. Nothing higher exists, indeed the concept of 'higher' does not exist here.

 

But if you take the same set of samples and dump them in a wider signal space (the purest way is inserting padding zeroes), you become aware of the images up to the limit of this new signal space. That is what oversampling on its own (*) is/does: move the samples to a wider space.

 

If you take a 44.1k sampled stream and move it to a 2.8MHz signal space, you will get the 0-22kHz baseband and its images up to 1.41MHz in that space. To get rid of the images one has to filter at 22.1kHz. Failing this one will not reconstruct the original signal. QED

 

 

Another approach is the direct application of the sampling theorem, which is proven, therefore truth. The theorem prescribes exactly one reconstructor which retrieves the original signal, not more, not less. This reconstructor is a specific filter at 22.1kHz. If by any means one would create from PCM samples a valid reconstructed signal, say by making a detour over DSD, then by implication this detour has to include the prescribed reconstructor, i.e. a filter at 22.1kHz. QED.

 

 

 

 

It is sad and worrying that I still have to explain basic things like these in 2014.

 

 

 

 

 

 

(* 'on its own' is key. In audio, oversampling is almost never done on its own, but always accompanied with the required reconstruction filtering.)

[Paul R]

Except.

 

Except what?

 

Say you have a signal sampled at 44kHz. It defines a spectrum from 0 to 22kHz. From 22kHz to infinite it defines a spectrum comprising of the 0-22kHz baseband positively and negatively mirrored around integer multiples of 44kHz. That is a fundamental mathematical fact.

 

If you regard the signal in its (limited) 44kHz sampled signal space all you are aware of is 0-22kHz. Nothing higher exists, indeed the concept of 'higher' does not exist here.

 

But if you take the same set of samples and dump them in a wider signal space (the purest way is inserting padding zeroes), you become aware of the images up to the limit of this new signal space. That is what oversampling on its own (*) is/does: move the samples to a wider space.

 

If you take a 44.1k sampled stream and move it to a 2.8MHz signal space, you will get the 0-22kHz baseband and its images up to 1.41MHz in that space. To get rid of the images one has to filter at 22.1kHz. Failing this one will not reconstruct the original signal. QED

 

 

Another approach is the direct application of the sampling theorem, which is proven, therefore truth. The theorem prescribes exactly one reconstructor which retrieves the original signal, not more, not less. This reconstructor is a specific filter at 22.1kHz. If by any means one would create from PCM samples a valid reconstructed signal, say by making a detour over DSD, then by implication this detour has to include the prescribed reconstructor, i.e. a filter at 22.1kHz. QED.

 

 

 

 

It is sad and worrying that I still have to explain basic things like these in 2014.

 

 

 

 

 

 

(* 'on its own' is key. In audio, oversampling is almost never done on its own, but always accompanied with the required reconstruction filtering.)

 

Except the second paragraph in the post you quoted?

 

What at is sad and worrisome is that you would put simple things in such a way as to spread FUD. A higher sample rate does not mean the signal must contain information above 22.1k, but it does mean you can use simpler less damaging filters. If you are math trained, this should be rather immediately apparent to you, and present a logical basis for a difference in sound.

 

As noted before, in the case of DSD, we are talking very simple filters indeed.

 

Your assertation that you still have to use filters while correct, is presented in a manner and tone that is entirely misleading to most people.

 

I won't comment on your attitude and assumption of authority in this area.

 

Paul

[Fokus]

A higher sample rate does not mean the signal must contain information above 22.1k, but it does mean you can use simpler less damaging filters.

 

NO!

 

If you put a 44.1kHz sampled signal in a 88.2kHz space, then that space WILL contain the 0-22kHz baseband, and then its first image from 22kHz to 44kHz. You only get rid of that image if you filter, hard, at 22kHz as part of the oversampling/repackaging exercise.

This is elementary.

[Jud]

 

It is sad and worrying that I still have to explain basic things like these in 2014.

 

 

(* 'on its own' is key. In audio, oversampling is almost never done on its own, but always accompanied with the required reconstruction filtering.)

 

You don't have to explain - unless you would like us to understand. For my part, I have been exposed only to the sort of "explanations" given by marketing people, plus a couple of very oblique references in this or other forums, so no, I have not learned even the basics. I have determined that I no longer wish to be so much in the dark, so I am going to try to teach myself something about the basic mathematics and filters. But meanwhile, if you are amenable, I would appreciate a brief explanation of the distinction (perhaps with an example from the typical digital recording-to-playback signal chain) you have mentioned between oversampling "on its own" and oversampling accompanied by the required reconstruction filtering.

[Paul R]

YES - it is not going to magically grow content above 22khz just because it is held in a different container. That is like saying a 3khz tone cannot be held in a 24/192k data file. Which is exactly what your words are implying.

 

Sample rate artifacts are a function of just that- sample rate. Not magically expanded content into a higher frequency. I grant that there are some special cases when upsampling redbook files, but they are for the most part, the same issues that would be present in a file recorded at high resolution. Just artificially created. Interpolation is the difference of course.

 

-Paul

 

 

NO!

 

If you put a 44.1kHz sampled signal in a 88.2kHz space, then that space WILL contain the 0-22kHz baseband, and then its first image from 22kHz to 44kHz. You only get rid of that image if you filter, hard, at 22kHz as part of the oversampling/repackaging exercise.

This is elementary.

[Fokus]

YES - it is not going to magically grow content above 22khz

 

The images above 22kHz are always implied, they are a firm property of a train of samples spaced 1/44100 seconds apart. These images grow real once the signal space is expanded, i.e. the spacing reduced, insofar the larger space encompasses the images.

 

Teresa was talking about converting CD to DSD. If you think that such is done without steep anti-imaging filtering at 22kHz then you are in for a massive disappointment. And now please think before you make an even greater fool of yourself.

[JohnSwenson]

Hi Guys,

Fokus is correct on this, let me try and explain it in a different way.

 

Think about what comes out of an R2R DAC, say a 1704. If there is no analog filter after the DAC chip, you get the infamous "stairstep" (which is actually a zero order hold of the actual data, but that is another topic). Voltage changes rapidly from one level to another at the boundaries of the sample time. In order for those sharp edges to occur there HAS to be higher frequencies involved. That is just basic Fourier. Without those higher frequencies you don't have the sharp edges. If you look at the output on a scope you see the sharp edges, so the high frequencies HAVE to be there.

 

Those higher frequencies are mirrored around the FS/2 (22.05KHz for 44.1). That means that if you have audio content at say 18KHz, it is ~4KHz from the FS/2 so it produces an "alias" at FS/2 + ~4KHz. Thus the higher the frequency of the original signal, the lower the frequency of the alias.

 

The only way to get rid of these "alias" signals is to filter them out. In the case of the 1704 being fed straight 44.1, it has to be done with a brickwall analog filter if you don't want to attenuate the 20KHz and down band.

 

You CAN do it with a digital filter, the output of which is a higher sample rate. Just converting to a higher sample rate without filtering doesn't actually change anything. Lets say you over sample to 88.2, but without filtering, what happens to the output of the 1704? Absolutely nothing! You get exactly the same stairstep. The only way it is going to change is if you perform filtering (DSP), this decreases the alias amplitude and "smooths" out the waveform. The output of the 1704 is still stairsteps (if you don't use analog filtering) but now each "step" is smaller and closer together. This means the aliases are lower in amplitude, and higher up in frequency. Now the aliases are mirrored around 44.1KHz and are lower in amplitude. This makes it much easier on an analog filter to "clean up" what is left. It doesn't have to be as steep to get reasonable attenuation.

 

Notice that in order for this alias attenuation to work well the digital filter has to do exactly the same thing as the analog filter did with the 44.1, it is the SAME brickwall filter. Well not completely the same because you can implement a whole range of digital filters that you can't implement with analog components, so it may actually be possible to implement a better filter with digital than analog.

 

Does this make any sense? I don't have access to any decent drawing program right now to make the pictures that go along with this, so just try and visualize it! (there are several places on the web that cover this exact info WITH nice pictures, I just can't find them right now)

 

John S.

[alfe]

You don't have to explain - unless you would like us to understand. For my part, I have been exposed only to the sort of "explanations" given by marketing people, plus a couple of very oblique references in this or other forums, so no, I have not learned even the basics. I have determined that I no longer wish to be so much in the dark, so I am going to try to teach myself something about the basic mathematics and filters. But meanwhile, if you are amenable, I would appreciate a brief explanation of the distinction (perhaps with an example from the typical digital recording-to-playback signal chain) you have mentioned between oversampling "on its own" and oversampling accompanied by the required reconstruction filtering.

 

Hi Jud

 

This can be helpful:

 

Upsampling vs. Oversampling for Digital Audio | Audioholics

[Jud]

You CAN do it with a digital filter, the output of which is a higher sample rate. Just converting to a higher sample rate without filtering doesn't actually change anything. Lets say you over sample to 88.2, but without filtering, what happens to the output of the 1704? Absolutely nothing! You get exactly the same stairstep. The only way it is going to change is if you perform filtering (DSP), this decreases the alias amplitude and "smooths" out the waveform. The output of the 1704 is still stairsteps (if you don't use analog filtering) but now each "step" is smaller and closer together. This means the aliases are lower in amplitude, and higher up in frequency. Now the aliases are mirrored around 44.1KHz and are lower in amplitude. This makes it much easier on an analog filter to "clean up" what is left. It doesn't have to be as steep to get reasonable attenuation.

 

Notice that in order for this alias attenuation to work well the digital filter has to do exactly the same thing as the analog filter did with the 44.1, it is the SAME brickwall filter. Well not completely the same because you can implement a whole range of digital filters that you can't implement with analog components, so it may actually be possible to implement a better filter with digital than analog.

 

John S.

 

Thanks, it is helpful. I think some of the confusion for me stemmed from the fact that I had always seen the terms "oversampling," "upsampling," "interpolation," etc., in conjunction with filtering, so I assumed the former automatically meant the latter. One thing: I could use some help in clarifying why the aliasing in the upsampling-plus-filtering example is now taking place around 44.1kHz rather than 22.05kHz. If you're not filtering, you're not changing the response, so the aliasing around 22.05kHz stays. If you *are* filtering, OK, the aliasing around 22.05 is ameliorated to whatever extent the filtering dictates. But does the aliasing at 44.1kHz when filtering is done to the upsampled bitstream come from the upsampling, the filtering, the interaction of the two, or none of the above?

[Jud]

Hi Jud

 

This can be helpful:

 

Upsampling vs. Oversampling for Digital Audio | Audioholics

 

I hope the article is better than the title! Hate stuff that tries to distinguish "upsampling" from "oversampling" when the two terms are by now so hopelessly confused that all you will get from them is which term the particular article's author wants to use, not something that is actually useful in wider discussion. (Sorry alfe, pet peeve of mine - thank you for helping me with the link.)

[alfe]

I hope the article is better than the title! Hate stuff that tries to distinguish "upsampling" from "oversampling" when the two terms are by now so hopelessly confused that all you will get from them is which term the particular article's author wants to use, not something that is actually useful in wider discussion. (Sorry alfe, pet peeve of mine - thank you for helping me with the link.)

 

The article have nothing to do with the title ¨-)