Page 1 of 3

44.1 kHz is slowly reaching its end

Posted: Sun Dec 17, 2017 4:11 pm
by Lyberta
I've noticed that both Blu Ray and UHD formats no longer have the option of 44.1 kHz sampling rate. This means that if you want your 44.1 kHz music to be used in such formats, you will need to resample which means some distortion.

Re: 44.1 kHz is slowly reaching its end

Posted: Sun Dec 17, 2017 4:16 pm
by nilshi
Good to know.
But also good to keep in mind that this is neither a technical problem for anyone nor a problem of convenience.

Re: 44.1 kHz is slowly reaching its end

Posted: Tue Dec 19, 2017 9:19 pm
by Luc
I believe I've read somewhere that 48 kHz has been the standard for video for a very long time. For audio (CD), 44.1 kHz still stands.

Re: 44.1 kHz is slowly reaching its end

Posted: Wed Dec 20, 2017 7:25 am
by Lyberta
Luc wrote:I believe I've read somewhere that 48 kHz has been the standard for video for a very long time. For audio (CD), 44.1 kHz still stands.
Well, both modern audio and video containers can hold 48 kHz. But it looks like modern video-oriented containers no longer support 44.1 kHz audio.

So 48 kHz simply has more support.

Re: 44.1 kHz is slowly reaching its end

Posted: Thu Jan 11, 2018 1:35 am
by loxstep
How much distortion is actually introduced when upsampling? Everytime I've had to do it, I haven't noticed any difference.

This video claims that there is no difference:
https://www.youtube.com/watch?v=ODWH5KanOu8

But I'm no expert. Either way, I doubt most people are listening to music on Blue Ray discs, unless it's part of a film score. Most people I know stream music, and I think CD audio standards still apply.

Re: 44.1 kHz is slowly reaching its end

Posted: Fri Jan 12, 2018 3:27 pm
by mclstr
loxstep wrote:How much distortion is actually introduced when upsampling? Everytime I've had to do it, I haven't noticed any difference.
There's no reason why you should be adding distortion when converting sampling speed in the digital domain.
It's not like in the analog world.

Re: 44.1 kHz is slowly reaching its end

Posted: Sat Jan 13, 2018 7:25 am
by Lyberta
loxstep wrote:How much distortion is actually introduced when upsampling? Everytime I've had to do it, I haven't noticed any difference.
Well, pretty much all floating point operations introduce small errors because floating point numbers have finite precision. Resampling from 44.1 kHz to 48 kHz means tons interpolation points that reside somewhere between the samples. So, in principle, there will be precision errors. But it's unlikely that they will be audible to the human ear.

Now, if interpolation is done in 64 bit, the errors will be inaudible by pretty much everything except super mega expensive science equipment.

Re: 44.1 kHz is slowly reaching its end

Posted: Sat Jan 13, 2018 7:37 pm
by Markus
To prove Lybertas statements I made some experiments.

I created a short bash script using sox (a high quality re-sampler) via ffmpeg to iterate a thousand times over a 3 second sample containing a 15 kHz sine wave of 44.1 kHz S/R. The script re-samples the last rendered sample to 48 kHz and back to 44.1 kHz in one iteration.

I then opened the original sample and the results of 1, 10, 100 and 1000 iterations in Audacity. I inverted the re-sampled results and mixed them with the original so only the difference of both samples is visualized.

Please bear in mind that one iteration actually means two re-sample processes 44.1 kHz -> 48 kHz -> 44.1 kHz.

The screenshot shows the four results in the order of 1, 10, 100 and 1000 iterations. To get an impression of the levels: 0.0001 equals ~-80dB, 0.001 equals ~-60dB. One can clearly see the steady values of the noise in the first two results indicating rounding errors. -80dB means the noise is below most analog hardwares own noise floor and in the range of quantization noise of the DAC.

The attachment resample.7z contains the original sample and the bash script. If you want to test it on your own make sure you ave ffmpeg with sox support installed ( ffmpeg 2>&1 | grep sox ). Download the archive and extract it to a new folder. cd into that folder and run ./resample [n] while [n] is the amount of iterations. Omitting [n] defaults to 100 iterations.

And here's the visual result:
resample_noise.png

Re: 44.1 kHz is slowly reaching its end

Posted: Sat Jan 13, 2018 10:48 pm
by folderol
Good work.
Not enough people actually test assumtions before accepting them.

Re: 44.1 kHz is slowly reaching its end

Posted: Sat Jan 13, 2018 10:59 pm
by Markus
After taking a closer look while thinking about it there's two thoughts. #1 If it was rounding errors there should be a pattern but there's none. So it might be #2 it's some white noise added by the re-sample process in order to prevent from artifacts or something.

Re: 44.1 kHz is slowly reaching its end

Posted: Mon Jan 15, 2018 12:20 pm
by Lyberta
Markus wrote:After taking a closer look while thinking about it there's two thoughts. #1 If it was rounding errors there should be a pattern but there's none. So it might be #2 it's some white noise added by the re-sample process in order to prevent from artifacts or something.
It's called dithering.

Re: 44.1 kHz is slowly reaching its end

Posted: Mon Jan 15, 2018 12:48 pm
by Markus
It's called dithering.
Which is common for bit-depth manipulation but not for sample-rate modifications?

Re: 44.1 kHz is slowly reaching its end

Posted: Tue Jan 16, 2018 7:26 am
by Lyberta
Markus wrote:Which is common for bit-depth manipulation but not for sample-rate modifications?
They are sort of the same. I first learned about dithering when reading about sample rate conversion.

Re: 44.1 kHz is slowly reaching its end

Posted: Thu Jan 18, 2018 12:16 pm
by Markus
Dithering is quite uncommon in sample rate conversions. Just some examples:

Adobe Audition:
https://helpx.adobe.com/audition/using/ ... types.html
No dithering in S/R conversion, but prominent in bit depth reduction.

iZotope:
http://downloads.izotope.com/guides/izo ... -ozone.pdf
The whole manual about dithering in iZotope is exclusively about bit depth conversion.

Voxengo:
http://www.voxengo.com/product/r8brainpro/
Again, dithering exclusively for bit depth conversion.

However. SOX tries to guess *) when to automatically enable dithering but one can force it to not use dithering at all. So I changed the script and ran the tests again. This time I went up to 10,000 iterations (20,000 conversions) so the order in the screenshot is 10, 100, 1000 and 10000 iterations. One can clearly see that more conversions don't introduce any more rounding errors after the first conversion, the results are exactly the same, which is the expected result.

Okay, enough time was burned, was just keen to see how much distortion is really introduced by S/R conversion(s). Never tested it on my own before and only relied on hearsay by now which was unsatisfying.

*) Update: Just to be complete, here's the reasons why SOX might enable dithering, taken from the man pages:

Code: Select all

· bit-depth reduction has been specified explicitly using a command-line option
· the output file format supports only bit-depths lower than that of the input file format
· an effect has increased effective bit-depth within the internal processing chain
resample.png

Re: 44.1 kHz is slowly reaching its end

Posted: Mon Jan 22, 2018 1:31 am
by loxstep
Markus - You are the man! :D
Great to see conventional wisdom put to the test. Now we know for sure that resampling isn't perfect.
Though I'm still not worried about it if the difference is inaudible.