Downsampling - The effects of ...........

[ISeekTheTruth]

Hi,
I am trying to get a grasp on the effects of downsampling audio. (And how to spot this in the spectrum for example) Specifically relating to what is sold commercially as CD quality. I guess it might fall into 3 camps
1) Downsampling same bit depth to a lower sampling rate. ex 48K to 44.1K
2) Downsampling a higher bit depth to a lower bit depth ex) 24 bit to 16 bit
3) Downsampling using a combination of #1 and #2
Any pointers to "good reading" or video would be appreciated. And again my root reason for looking into this is to understand and be able to spot poor/good quality commercially sold "CD Quality" audio content. Thanks

 

[RayDunzl]

I'm no expert, but...

1) Downsampling same bit depth to a lower sampling rate. ex 48K to 44.1K

There would have to be a mathematical interpolation of the sample values since the new samples won't occur at the same time

2) Downsampling a higher bit depth to a lower bit depth ex) 24 bit to 16 bit

Here, the lowest 8 bits lost are all subdivisions of the smallest of the remaining 16 bits. They might round the result up or down, or some other caclulation (maybe based on rate of change), and dither, and its done.

3) Downsampling using a combination of #1 and #2

If a competent resampler is used there shouldn't be a problem in either case.

---

Click though the list of converters here (click the arrows next to "sweep", for some reason) and see some examples of potential artifacts.

SRC Comparisons

src.infinitewave.ca

 

[DVDdoug]

CDs are 16-bit, 44.1kHz, 2-channel stereo*, and uncompressed PCM (same underlying format as WAV) so that's "CD quality" and there are no options/variations.

The modern "studio standard" is 24-bit, 96kHz and CDs (and MP3s, etc.) are downsampled from that.

The lossy formats don't store individual samples so they don't have a "bit depth" (and MP3 has more dynamic range than CDs). MP3s are limited to 48kHz and everything I've downloaded from Amazon has been 44.1kHz. Other lossy formats can have higher sample rates.

And again my root reason for looking into this is to understand and be able to spot poor/good quality commercially sold "CD Quality" audio content.

There a LOT of things that affect sound quality so even a 24/96 recording can "sound bad". There are rumors of CDs made from MP3 (although that's rare with commercial CDs from the big record companies). It's more common to have "fake" high-resolution files up-sampled from CDs. And of course there are many older recordings from digitized analog. Even some newer recordings are done on analog tape, or there are mastering engineers who use analog processors.

The folks at HydrogenAudio who do ABX Tests have pretty-much concluded that you can't hear the difference between a high-resolution original and a copy down-sampled to CD quality (under "normal" or "reasonable" listening conditions in a proper-scientific, level matched, blind listening test).

I've converted between CD-quality and 16/48 for video in both directions (and I never use anything less than CD-quality). I've never heard any defects or side-effects. And I've never tried to hear a difference with an A/B or ABX test... If I have to make a conversion there's no sense in worrying about it.

I also have lots of MP3s that I've ripped from CDs (at LAME V0 which is the "best" variable bitrate setting). The MP3s are loaded onto an iPod that's "permanently" docked to my car stereo. Again, I haven't tried to hear a difference and every time I've thought I was hearing a compression artifact, the same "defect" exists on the original CD.

 

[ISeekTheTruth]

For # 2 It is interesting for 24 bit they tacked on the 8 bits on the least significant side of the 16 as I would not think it would give them much more dynamic range. (Being suborders of the smallest 16 bit increment) I had heard dynamic range increases with bit depth. I do not know, just thinking about it. And again as you say, the conversion with the 8 bits on the lower order side would just require stripping them off and you would have a resultant 16 bit sample accurate to the smallest increment. Dithering.. I have heard the term. will read a bit. Thanks.

 

[ISeekTheTruth]

Doug - As you say, hearing the difference is what matters at the end of the day. I just always worry about hidden anomalies in conversion that surface under specific conditions. Like Amir uses a specific test file that brings out the worst in the different types of equipment he tests.

 

[RayDunzl]

Dithering.. I have heard the term. will read a bit.

Dither...

Green - 1kHz tone, 24 bit (high precision, but still has some "error" at very low levels because - well, the last 0 or 1 should often be some fraction of 1 to accurately describe the sine. The error is over 150dB down, so, who cares.

Red - 1kHz tone, 16 bit (lower precision) has more error in the last bit. A little over 100dB down, so, could be a problum in edge cases.

Black - 1kHz tone, 16 bit, with dither (some randomness added to/subtracted from the 16th bit), to cloak those nasty tones you can't hear in noise you can't hear so it looks better.

Dither is typically added when converting from one sample rate to another.

For recordings of real music, noise in the recording process (room, tape, people breathing, traffic, mic noise, etc) swamps all of that, with that noise level maybe around -50 or -60dB.

 

[RayDunzl]

Green - 1kHz tone, 24 bit (high precision, but still has some "error" at very low levels because - well, the last 0 or 1 should often be some fraction of 1 to accurately describe the sine. The error is over 150dB down, so, who cares.

Ok, maybe somebody cares.

Add blue, the dithered 24 bit tone:

 

[ISeekTheTruth]

Nothing like something visual to clarify things.. Really appreciate that. As to my previous comment on #2, it Would make more sense to paste the new bit width on the least significant side to accommodate the up and coming higher sample rates. The dynamic range increase would more of a bonus.

 

[ISeekTheTruth]

So to recap. Downsampling should not be a problem and the resultant file should not exhibit issues other than (Pending sanctioned conversion utilities are used)
- capping of the dynamic range that might exist in the source file
- capping of the available high end frequencies that are in the source file
- both of the above would not be exhibited in any way as distortions in the resultant audio file.
Or is any remastering required when a 24-bit, 96kHz is downsampled to 16 - 44.1K?

 

[RayDunzl]

Or is any remastering required when a 24-bit, 96kHz is downsampled to 16 - 44.1K?

You could convert a file on your own and listen to the before and after renditions...

Good luck hearing a difference.

If you need a file to play with, http://www.2l.no/hires/index.html? has many formats from which to choose.

 

[ISeekTheTruth]

My mother was a porpoise.. Good url references as always. Thanks again.

 

[Apesbrain]

I routinely convert 24/96 to 16/44 -- no "remastering"/EQ involved -- and do not hear any difference. Easy enough to test this yourself.

 

[earlevel]

Hi,
I am trying to get a grasp on the effects of downsampling audio. (And how to spot this in the spectrum for example) Specifically relating to what is sold commercially as CD quality. I guess it might fall into 3 camps
1) Downsampling same bit depth to a lower sampling rate. ex 48K to 44.1K
2) Downsampling a higher bit depth to a lower bit depth ex) 24 bit to 16 bit
3) Downsampling using a combination of #1 and #2
Any pointers to "good reading" or video would be appreciated. And again my root reason for looking into this is to understand and be able to spot poor/good quality commercially sold "CD Quality" audio content. Thanks

Some good replies already, so I'm just starting with your original request and adding some subtleties...

First, reducing bit depth is not down sampling, it's truncation or bit reduction. This can result in distortion that's dependent on what the signal is doing. Dither breaks that dependence on the signal—a low level hiss is far less distracting than a low level scratching that comes and goes.

Here is the bottom line of bit depth: If dithered, the 24-bit original and the 16-bit dithered version will sound exactly the same, except that the 16-bit version will have a higher noise floor. In fact, if you subtract one from the other, you'll end up with nothing but the background hiss (pretty quiet at 16-bit, you'll have to turn up your system to hear it). The conversation gets slightly more complicated if you're talking about noise shaping, but is fundamentally the same.

I could point you to info on the rate conversion, it's fundamentally a matter of filtering quality, and not hard to get results that are better than your ear's ability to tell. Dither is extremely simple to understand, so if you want to understand it as well as anyone, this will cost you just a few minutes:

This gives you a little better idea of exactly what dither fixes, with examples:

 

[ISeekTheTruth]

Great. And I DO appreciate being corrected on terminology.

 

[ISeekTheTruth]

Excellent videos. It will take a bit of time to get it jelled in my brain (Naked truth one), but I can see there is strong content with alot of practical examples. Time for a salmon sandwich (Brain food). Thanks.

 

[ISeekTheTruth]

Salmon sandwich did it. Clear now for bit reduction and the cost of noise from dithering wrt resultant bit depth. This ended up surfacing the topic of where do we end regarding bit depth and sample rates. Dither is an example when you get the into higher bit depths, the tradeoff noise is essentially inaudible. And sample rates must have their audible limits as well. Then the up and down side of the higher sample rates. Aliasing? Feels like I am peeking down a rabbit hole.....

 

[ISeekTheTruth]

Just thinking if I used a good source test file I could use Reaper to look first hand at much of what I have been researching. Any pointers to good free or reasonably priced plugins to look at what we have been discussing? Then I will leave you guys alone. Really have appreciated the great sources of information. A great start into diving into digital.

 

[earlevel]

Salmon sandwich did it. Clear now for bit reduction and the cost of noise from dithering wrt resultant bit depth. This ended up surfacing the topic of where do we end regarding bit depth and sample rates. Dither is an example when you get the into higher bit depths, the tradeoff noise is essentially inaudible. And sample rates must have their audible limits as well. Then the up and down side of the higher sample rates. Aliasing? Feels like I am peeking down a rabbit hole.....

Right, no only do the bits quickly outstrip our hearing, but also the electronics—24 is more that we'll ever hear or reproduce with electronics, so we're set for life at 24.

Many threads on the sample rate issue, here and other boards, endlessly argued. Personally, I don't need the higher sample rates. There is true value for things like non-linear DSP (typically in the recording stage, not home audio)—distortion, compression, etc.—but that is handled just fine by rate conversion within the process (plugin, etc.) itself. But why start an old tired argument here

 

[ISeekTheTruth]

As you have surmised, I'm in the home audio camp. I do have a hobby in recording but is limited to low volume ambience type clips. Raw with minimal filtering and no effects. Noise floor, wind and airplanes are my biggest enemies.

 

[ISeekTheTruth]

One last clarification. on dither and tearing. I am seeing posts that each bit ? has 6dB of dynamic range. There must be some context to this blanket statement.