Windows resampling not actually that bad?

[Degru]

I saw the awful measurements of Windows 10 resampling on archimago's site here: http://archimago.blogspot.com/2015/11/measurements-windows-10-audio-stack.html

and I figured I would test this for myself. I see quite a few people pointing to that article when they say Windows resampling sucks, and I feel like results like that should be very audible, whereas i don't really hear that big of a difference if at all. I only have a crappy laptop ADC, but it should still be able to capture the awful results shown in the article.

Setup is Geekout 1000 connected to one laptop with the standard Windows UAC2 audio driver and latest updates installed, outputting 19+20khz sines via ARTA through DirectSound at various samplerates, then running the output through my Fiio A5 to attenuate the signal, into the ADC of my other laptop which is recording the result at 96khz and displaying in ARTA.

Well, behold. 44 -> 96khz resampling through Windows.

Nothing even close to the results in the article, tho not spectacular either. Keep in mind that most of these spikes are caused by the craptastic ADC rather than the resampler. Here is non-resampled output for reference:

I did a bunch of other combinations of samplerates as well, attached to this post. Just about the only results that were meaningfully worse were the 44->48 and 48->44 tests, IMO.

User Case on Hydrogenaudio got similar results here: https://hydrogenaud.io/index.php/topic,114138.msg940216.html#msg940216

In conclusion, I think there is nothing to worry about with Windows resampling.

 

[Ron Texas]

Never mind, I will find something to worry about. Thanks for the heads up.

 

[PaulD]

This site has many SRC implementations tested, including Win 10 http://src.infinitewave.ca

 

[solderdude]

From http://src.infinitewave.ca we can read that the resampler spurious are around -110dB.

The rest of the errors remains below -130dB so will be masked by the measurements as the noise floor is higher (-120dB).
That doesn't mean it isn't there.
Personally I don't think I can hear that but for the sake of excellence I would look for a better resampler like the freeware Audicity 2.0.3. best Q for instance or SoX linear phase

 

[Degru]

From http://src.infinitewave.ca we can read that the resampler spurious are around -110dB.

View attachment 34091

The rest of the errors remains below -130dB so will be masked by the measurements as the noise floor is higher (-120dB).
That doesn't mean it isn't there.
Personally I don't think I can hear that but for the sake of excellence I would look for a better resampler like the freeware Audicity 2.0.3. best Q for instance or SoX linear phase

Well, can't exactly use other resamplers in DirectSound. I use SoX linear to 44100 for my non-44100 music in foobar, and have Windows set to 44100.

 

[dc655321]

It should be noted Archimago's tests were published 4 years ago. A lot can change in software over that time span.

Perhaps you could ask him to follow up with a repeat to confirm your findings?

 

[Vincent Kars]

Doesn't looks good to me.

 

[August]

In order to avoid this kind of degradation of sound quality, I will use wasapi and ASIO to play music.

 

[Fluffy]

It seems there is a contradiction between different measurements. I don't think we can draw any conclusion until we can reconcile this difference.

 

[ernestcarl]

Yep. Doesn't look good. But I think I'll continue to use DS for the sake of convenience. At my my calibrated maximum reference volume of 74 dB, I don't think it makes that much of a noticeable difference. You're mileage may vary.

 

[bennetng]

View attachment 36965

Doesn't looks good to me.

It has nothing to do with the resampler. See the video from 3:40 to understand why.

It is caused by CAudioLimiter APO:
https://matthewvaneerde.wordpress.c...audiometerinformation-and-full-scale-signals/

 

[pkane]

I saw the awful measurements of Windows 10 resampling on archimago's site here: http://archimago.blogspot.com/2015/11/measurements-windows-10-audio-stack.html

and I figured I would test this for myself. I see quite a few people pointing to that article when they say Windows resampling sucks, and I feel like results like that should be very audible, whereas i don't really hear that big of a difference if at all. I only have a crappy laptop ADC, but it should still be able to capture the awful results shown in the article.

Setup is Geekout 1000 connected to one laptop with the standard Windows UAC2 audio driver and latest updates installed, outputting 19+20khz sines via ARTA through DirectSound at various samplerates, then running the output through my Fiio A5 to attenuate the signal, into the ADC of my other laptop which is recording the result at 96khz and displaying in ARTA.

Well, behold. 44 -> 96khz resampling through Windows.
View attachment 34073
Nothing even close to the results in the article, tho not spectacular either. Keep in mind that most of these spikes are caused by the craptastic ADC rather than the resampler. Here is non-resampled output for reference:
View attachment 34074
I did a bunch of other combinations of samplerates as well, attached to this post. Just about the only results that were meaningfully worse were the 44->48 and 48->44 tests, IMO.

User Case on Hydrogenaudio got similar results here: https://hydrogenaud.io/index.php/topic,114138.msg940216.html#msg940216

In conclusion, I think there is nothing to worry about with Windows resampling.

I found some issues with aliasing in the Windows 8 resampler. Most likely not audible, but definitely measurable. For what I was using it for, this was significant enough for me to go ahead and write my own resampler


https://www.audiosciencereview.com/...ed-by-peter-pawłowski.6452/page-3#post-230325

 

[Vincent Kars]

It is caused by CAudioLimiter APO

Even at -60 dBFS?

 

[bennetng]

Even at -60 dBFS?

If the input signal is higher than that 0.985 thing, even if you reduce the Windows mixer volume to -60dB artifacts still exist. However, you can see that I reduced foobar's preamp just by 0.2dB before leaving foobar, then there is no artifact with subsequent volume reduction.

For this reason, there is also a WASAPI shared plugin:
https://www.foobar2000.org/components/view/foo_out_wasapis

With this plugin, volume can be reduced by using the standard volume control without artifact while maintaining shared mode behaviour.

 

[ofrappier]

Hi,

A dream : that Windows use only one good AUDIO core : like CORE AUDIO of OSX
(see measurement of OSX core audio in another post)

Regards,

O.

 

[bennetng]

Some professional software...

 

[ofrappier]

and OSX vs WIN :

And coreAudio have been improved since OSX Leopard.....

 

[bennetng]

Keep in mind that most of these spikes are caused by the craptastic ADC

IMO the ADC seems not bad at all, the question is price. What interface do you use?

 

[Degru]

IMO the ADC seems not bad at all, the question is price. What interface do you use?

This is actually just the audio input built into my Thinkpad x201s. I have a Dell Precision M4600 with a cleaner input, but it is really designed for mics and can't handle line inputs properly. The Thinkpad input is a lot better behaved in that regard, plus it's much more portable and easy to set up.

 

[bennetng]

Okay here is MPC-HC without using exclusive mode, playing a 76543Hz sweep file at 16 and 24-bit, recorded at 24-bit 48kHz using digital loopback. Left hand side is 100% volume and right hand side is 95% volume.

In terms of frequency...

and amplitude.

As you can see the test file's right channel is at 50% volume (-6dB) but it still distorts when using 100% playback volume. Just slightly reduce the built-in volume control in MPC-HC eliminates these artifacts and you can see how clean the Windows SRC is. When playing the 16-bit test file the noise floor is waaaaay higher than everything else. That means if Windows SRC is really that bad then everyone should delete all of their 16-bit files and sell all of their CDs.