My quest for Nyquist

All Things Music
1

The great debate. Are higher sampling rates important? Last night I listened to a FLAC copy of “So What” by Miles Davis through my NAD Viso HP50’s and my Dragonfly Black. I’ve listened to this track hundreds of times with a wide variety of headphones, but this was the first time I listened to the hi-res version. I could definitely tell he difference.

Then I watched some YouTube videos over whether or not we only need to engineer to the limits of human hearing ala the Nyquist theorem. So, where do you stand on this? Is digital music sampled at 48kHz worse than digital music sampled at 96 or 192kHz?

3 Likes
2

This is a great topic. Would love to hear everyone’s thoughts.

3

In my experience, it is more likely by playing hi-res through the Dragonfly, you are bypassing some of the digital filtering, which oversamples the music internally before converting it to analog. The digital filters on DAC chips are pretty basic so high-res can sound better in that case, though it’s generally only really noticeable on good, minimally processed acoustic recordings.

The other, and far more likely possibility, given the history of Kind of Blue is that the “CD Quality” version you listened used a different mastering to the hi-res version. The original versions were known to have issues (mainly with the pitch) and there have been various attempts to fix those issues with remasters. David Chesky finally managed to get hold of the original analog tapes and completely remaster the album from scratch. If that is the version you listened to in hi-res then it does indeed sound different to the older CD versions as they fixed all the issues as well as being able to extract more audible detail.

6 Likes
4

I am listening to the “Legacy” version of the track. I have tried the following combinations:
Music: Apple Music
Source: iPhone 7
DAC: Dragonfly Black
Headphones: NAD VISO HD50, Sennheiser PX 200, Grado SR60

Music: Tidal
Source: iPhone 7
DAC: Dragonfly Black
Headphones: NAD VISO HD50, Sennheiser PX 200, Grado SR60

Music: FLAC via Vox player app
Source: iPhone 7
DAC: Dragonfly Black
Headphones: NAD VISO HD50, Sennheiser PX 200, Grado SR60

Music: Apple Music, Tidal, Vox
Source: iPhone 7
Headphones: Sony MDR1000x (Bluetooth, N/C on)

In all cases, the higher resolution FLAC version of “So What” sounded clearer, more vibrant, better separation. This version was much better at highlighting the headphones’ strengths and weaknesses.

I’ve got to add that I’m no expert and my gear is pretty basic. I also spent hours in front of large speakers at rock concerts so my hearing isn’t that great. But I can still distinguish between these versions.

5

Per @Currawong, in most cases finding a readily audible difference between a standard CD-resolution file (16/44.1) and a commercially available high-resolution copy primarily tends to be a) down to them coming from different masters (most common) or b) because your’e bypassing/using different filtering in your DAC*.

There’s a c) case as well … which involves source material that has a lot of ultrasonic content played back on gear that doesn’t have the bandwidth to handle (or filter) it properly - and that can result in all sorts of undesirable audible artifacts/effects (and might not be good for the gear in question, either).

The only way to be sure that you’re not getting different masters is to start with a high-resolution file and then create lower resolution versions yourself, using high-quality re-sampling tools (e.g. iZotope or SoX) to produce the lower resolution copies from that common source file**. Any other test material is a bit too open to other factors to be trustworthy.

You can then use something like Foobar with the AB/X module to perform a controlled, blind, test and see where you wind up. It’ll even log the outcome for you. This tends to be quite illuminating and is worth trying for yourself!

I’ve performed this sort of test myself a number of times.

The vast majority of the time I cannot reliably discern any difference between the 16/44.1 down-sampled file and the original 24/192 (or higher) original. This is with short fragments of each track for each comparison iteration and a DAC that doesn’t oversample/filter differently based on bit-rate.

With a DAC that does oversample and/or have different filter behavior with changing input rates I can, in general, tell a difference, with reliability better than statistical chance. But I have to be focusing on what I’m hearing to a sufficient degree that I would never actually listen to music that way.

Finally, I’ve done longer tests, with entire albums processed this way, and the version I’m listening to selected at random (hardware random number generator), and typically found that I’m less “listening fatigued” at the end of my listening that I was with the lower-resolution content. But there are so many variables at work with this sort of comparison, that it’s a tenuous correlation at best.

Now, do I buy high-resolution content?

Yes.

But only either a) when I have knowledge of/provenance for there being a material difference in the masters used for the high-resolution version and the commonly available CD or streamed version (e.g. the “Kind of Blue” reference @Currawong makes above) or b) it’s the only way to buy the album for download.

TL;DR; - I don’t doubt that using different sources for the different resolutions results in distinctly audible results - but the source resolution is probably not, by itself, the immediate cause. Where source resolution difference is the only variable on the input, audible differences are likely to be down to different/bypassed DAC filtering or other equipment artifacts.

*One way around this is to take the downsampled file, and then upsample it back to a level where your DAC’s filtering/upsampling is not in effect.
**This needs to be done properly and requires either quality tools with proper pre-sets, or you need to understand what you’re doing so you can apply appropriate dither when reducing bit-depth etc.

6 Likes
6

This is my favorite article on the matter:
https://people.xiph.org/~xiphmont/demo/neil-young.html

I personally took several of these high sample rate flacs and converted them to -

  1. 44.1khz flacs
  2. 44.1khz, 320kbps mp3
  3. 44.1khz, 128kbps mp3

I then proceeded to blind test the different versions of the songs using Foobar2000’s ABX Comparator plugin.

Across the board, I couldn’t tell the difference between the original flacs, 1 and 2. Only 3 had noticeable differences to my ears.

This is just my personal experience though. I’m not dismissing the possibility that someone with better ears/gear/experience would notice differences that I might have missed.

1 Like
7

Wouldn’t the identical sample rate in your test make the differences less noticeable? That and two of the three are MP3 compression.

It would seem that a higher sampling rate would lead to a closer match to the original recording, assuming there was a good recording to begin with. And that, of course, would depend on the technique used to capture it.

I wonder what a blind test of the same recording at 192, 96, 48, and 44.1 kHz would show. I know sound quality is a lot more complex than this, but this seems to be the core of some debates about streaming and downloadable music.

8

Sorry in case it wasn’t clear enough in my original post - I was comparing 4 versions of each song, not 3. I compared the 3 versions to the original, high sample rate flac.

9

In most cases it would show that higher resolution capture tends to exhibit fewer artifacts resulting from imperfect brick-wall and anti-imaging filters. Which is largely because those effects become smaller with regard to the energy present in the frequencies that occur in actual music as capture-time sampling frequency increases…

If you replayed the original recordings, in their original capture format, you might be able to hear the results of the artifacts* that are present from imperfect filtering at lower capture rates.

However, studios generally capture at higher sampling rates to work around this specific issue. And once that capture has been done at the higher rate, downsampling to create a “lower resolution” copy of that source does not undo the benefits of that higher-rate capture.

So, absent there actually being musical information above the sample rate of the eventual container, or the dynamic range of the original material being greater than the format it is converted to, no information is lost in the conversion from, say, 24/192 to 16/44.1.

Now, if you needed 17 bits per sample to cover the full dynamic range of the source, that would be different. As it would be if there was signal content in the source above 22,050 Hz. Whether you could then hear that loss is another matter, but it’s unlikely.

*These artifacts occur when there is energy present in frequencies above half the sampling frequency are sampled. The brick wall filter is supposed to stop those frequencies from reaching the ADC, but such filters aren’t perfect. The images that result occur at LOWER frequencies (their called images or aliasing artifacts), and are within the audio band even if the cause of them is not).

3 Likes
10

What are some examples of the artifacts that you might hear at 44.1 kHz vs. higher sampling rates?

11

Essentially, false signal at fractions of the sampled source frequency. In extreme cases you’d get a false “tone”.

But it’s not about 44.1 kHz vs. higher frequencies specifically. It’s about the relationship to frequencies in the source material vs. the sampling rate.

Start with this basic discussion of anti-aliasing filters and you can do more reading from there.

You still have to have content in the source at higher frequencies than the half the sample rate of the ADC actually reaching the ADC for this to happen at all.

1 Like
12

@Torq and @alont really nailed it with the technical considerations when comparing hi-resolution files vs lower sampling rates. Their suggested method of starting with the highest sampling rate file, and then using that source file to make your own lower sampled versions, and then using blind testing like Foobar’s ABX comparator is the only way to be sure you are comparing apples to apples. Trying to make any type of comparisons across different sources like streaming sites or CDs will not be sufficient due to the different masters being used.

To emphasize this point about considering the masters being used, there was a very good reddit post showing examples of how to quantify the quality of a master by using a software called AudioLeak, which uses the method Leq(A) to show a spectrograph of the A-weighted loudness of a track across time. The author of the post does a great job in explaining this more and provides some great examples of what this actually looks like. When considering the same song, different mastered versions can have drastically different qualities in production. From the author’s photo album, here is an example of “New Religion” by Duran Duran:

Original recording (1982):

evK4767.jpg806×328 59 KB

Remaster (2009):

ZXGy4lF.jpg1046×328 70.9 KB

It is clear that the remaster version is clipped and of shows a much worse dynamic range compared to the original recording. One of the issues with streaming sites as well as digital music stores is that it is sometimes nearly impossible to track down exactly which master version they are providing regardless of how they actually label it on their site. This is another reason that comparing the different streaming sites to each other is so difficult, often times each of the different services is using a slightly different master before doing their lossy conversion.

4 Likes
13

It’s probably incredibly difficult for most people to detect these aliased images even if you had a slow roll-off low pass filter - especially in music when you have masking effects in play. If this is a concern, there is a test tone on audiocheck.net that lets you hear how ‘leaky’ your low pass filter is. If you don’t detect any spurious tones before the main sweeping sine tone, then the filter is doing its job.

14

I wouldn’t argue to the contrary.

I’m simply pointing out that recording at higher sample rates than double the maximum frequency present in the source material is only useful to offset the potential for imaging issues (audible or not) resulting either a) from the lack of a perfect brick-wall filter (i.e. always), or b) when you have insufficient frequency headroom at a lower sample rate to allow a more practical, low-pass/roll-off filter, of a more typical 6-12 dB/octave.

In other words, a higher sample rate will not capture the lower frequency information with any more fidelity than a lower sample rate, provided you have no energy in frequencies above 0.5 fs of that sample rate reaching the quantizer.

1 Like
15

@alont

That makes more sense. Maybe it is that the recording I’m listening to was a better master as some have suggested.

It also sounds (not a pun) that the combinations of master, encoding, playback device, amp/DAC, and headphones are essentially infinite. Add to that the huge variation in hearing capacity and personal taste.

Before you know it, it’s a hobby with a discussion board. Oh, wait. It is.

16

That would be my guess as well.

17

I think you are all correct. I’ve been listening to streaming versions of some of the classics for so long, I forgot what CD quality really means. The Miles Davis CD was the “Legacy” edition that was recently remastered.

Now I’m wondering if I need to get a little CD player again, hook it up to a good amp and make myself a critical listening rig.

1 Like
18

I don’t know what CD quality is on Miles’ Kind of Blue. I have vinyl.

19

Original or remastered?

20

The other, and far more likely possibility, given the history of Kind of Blue is that the “CD Quality” version you listened used a different mastering to the hi-res version.

I would guess the same. One easy way to control for this @jflores476 would be to convert the hi-res version to 16/44.1 and then see how that sounds. If you’re feeling really adventurous, you could use some ABX testing software to easily compare them side-by-side.

I actually just bought a DSD version of Michael Jackson’s “Thriller”, not because I think I can tell the difference between DSD and Redbook PCM, but because this comes from the SACD mastering of that album which seems to be highly regarded.

In my imaginary dream world, I have access to a music catalog that allows me to pick any mastering of any album and download it at any bit depth and sampling rate.

2 Likes