72 software Sample Rate Converters put to the test

[John Kenny]

Sorry, here it is the original plots:
https://dropfile.to/msk45

As far as the "missing" colors, they are perfectly superimposed plots.

Thanks for the link
Yea, I got that they were perfectly superimposed but it's knowing which plots they are superimposed on is where the confusion lies - some plots are "bad" & some "good" on the same graph

 

[bibo01]

Thanks for the link
Yea, I got that they were perfectly superimposed but it's knowing which plots they are superimposed on is where the confusion lies - some plots are "bad" & some "good" on the same graph

In each plots colors are always the same:
GREEN
RED
PURPLE
YELLOW
BLUE
WHITE

If you cannot see some of them, it's because they are perfectly superimposed. Usually the missing ones belong to the same family brand.

 

[John Kenny]

In each plots colors are always the same:
GREEN
RED
PURPLE
YELLOW
BLUE
WHITE

If you cannot see some of them, it's because they are perfectly superimposed. Usually the missing ones belong to the same family brand.

I understand that but you don't understand what I'm saying - let me give you an example:
In this graph what we see are green, red & white plots
So which of the unseen plots (yellow, blue, purple) overlay the green plot - which overlay the red plot, etc?
I'm presuming blue, yellow & purple all overlay the white plot but I'm not sure?


GREEN = AUI Converter v5.6.3 free - No Dither
RED = AVS Audio Converter v8.1
PURPLE = AVS Audio Editor v8.1 - colorized dither, noise shaping A
YELLOW = AVS Audio Editor v8.1 - colorized dither, noise shaping B
BLUE = AVS Audio Editor v8.1 - gaussian dither, noise shaping A
WHITE = AVS Audio Editor v8.1 - gaussian dither, noise shaping B

P.S.: unseen colors are superimposed

Again in the next plot, I'm not sure where green, red, purple & yellow overlay?

GREEN = AVS Audio Editor v8.1 - Triangular dither, noise shaping A
RED = AVS Audio Editor v8.1 - Triangular dither, noise shaping B
PURPLE = AVS Audio Editor v8.1 - uniform dither, noise shaping A
YELLOW = AVS Audio Editor v8.1 - uniform dither, noise shaping B
BLUE = AVS Audio Editor v8.1 - no dither
WHITE = Awave Studio 10.6 - FIR filter

 

[bibo01]

I understand that but you don't understand what I'm saying - let me give you an example:
In this graph what we see are green, red & white plots
So which of the unseen plots (yellow, blue, purple) overlay the green plot - which overlay the red plot, etc?
I'm presuming blue, yellow & purple all overlay the white plot but I'm not sure?


GREEN = AUI Converter v5.6.3 free - No Dither
RED = AVS Audio Converter v8.1
PURPLE = AVS Audio Editor v8.1 - colorized dither, noise shaping A
YELLOW = AVS Audio Editor v8.1 - colorized dither, noise shaping B
BLUE = AVS Audio Editor v8.1 - gaussian dither, noise shaping A
WHITE = AVS Audio Editor v8.1 - gaussian dither, noise shaping B

P.S.: unseen colors are superimposed

Again in the next plot, I'm not sure where green, red, purple & yellow overlay?

GREEN = AVS Audio Editor v8.1 - Triangular dither, noise shaping A
RED = AVS Audio Editor v8.1 - Triangular dither, noise shaping B
PURPLE = AVS Audio Editor v8.1 - uniform dither, noise shaping A
YELLOW = AVS Audio Editor v8.1 - uniform dither, noise shaping B
BLUE = AVS Audio Editor v8.1 - no dither
WHITE = Awave Studio 10.6 - FIR filter

In the first plot PURPLE-YELLOW-BLUE overlay WHITE
In the second plot GREEN-RED-PURPLE-YELLOW overlay BLUE

 

[John Kenny]

In the first plot PURPLE-YELLOW-BLUE overlay WHITE
In the second plot GREEN-RED-PURPLE-YELLOW overlay BLUE

Thanks!

 

[amirm]

I might need a tutor before hand BTW did you figure out how to test that junk I sent you yet?

Not yet. Been too busy with this forum I started....

 

[Isabella Lee]

There is also a great Audio Converter for Windows for you to test. This professional software can work at a wide variety of audio formats.

Mod edit: That software is for removal of copy protection from online content. I have edited out the link. If this is not spam, you can follow up by explaining how it relates to work we do here.

 

[bennetng]

Just want to point out null tests will highly favour linear phase resamplers, minimum/intermediate phase settings will screw up the results. Also the potential clipping issue if the conversion and comparison is not being done in floating point.

infinitewave's methodology is more sensible.

 

[Ron Texas]

Could someone give me a summary for dummies? Is there a preferred algorithm or setting that can be recommended? There is a lot of data here and I don't know how to read the charts.

 

[bibo01]

Just want to point out null tests will highly favour linear phase resamplers, minimum/intermediate phase settings will screw up the results. Also the potential clipping issue if the conversion and comparison is not being done in floating point.

infinitewave's methodology is more sensible.

That is true in both instances.
The null test, if with a minimum phase, involves a greater difficulty of alignment, moreover some resamplers introduce phase rotations at low levels.
If the signal is near 0dBFS, in some cases some samples are shaved off.
There are obviously remedies, so every situation should be verified as it was in this test.

 

[bennetng]

Some people are also skeptical about iZotope's results in infinitewave's website since Alexey Lukin from iZotope is involved with test design.

As a third party test, your results do show that iZotope's SRC is really superb in quality. For linear phase resamplers without any subsample delay and phase change, null test is indeed a very good method to evaluate SRC quality.

 

[Kees de Visser]

As a third party test, your results do show that iZotope's SRC is really superb in quality. For linear phase resamplers without any subsample delay and phase change, null test is indeed a very good method to evaluate SRC quality.

AFAIK most SRCs don't really care about subsample delay (it has no influence on sound quality). iZotope changed the code (Alexey told me) to avoid subsample delay after requests by users who do null tests (like me). It's just cosmetics, but it makes them look better in tests like this.
IMO null tests are meaningless, unless there is no subsample delay (and phase difference).

 

[bennetng]

Totally agree.

 

[daftcombo]

Do you know how foobar resampler would do on this test?

 

[daftcombo]

This comparator has already been posted elsewhere on the forum but it should appear in this thread:

https://src.infinitewave.ca/

 

[daftcombo]

I wanted to compare Foobar2000 SoX resampler + bit rate reducer VS the same functions in dBpoweramp.

Pretty big difference on a track imho.

Here's the original first:

And the converted files:

 

[bennetng]

I wanted to compare Foobar2000 SoX resampler + bit rate reducer VS the same functions in dBpoweramp.

Pretty big difference on a track imho.

Here's the original first:

View attachment 61380

And the converted files:

View attachment 61381

foobar's built-in dither is always noise-shaped, and high frequency noise is caused by the noise-shaped dithering algorithm rather than the SoX plugin.

You can use the dither plugins here for non-shaped dither, and turn off foobar's built-in dither.
https://foobar.hyv.fi/

Search for the word "Intuitively" in this article to see a proper way to evaluate resamplers.
https://archimago.blogspot.com/2019/06/guest-post-why-we-should-use-software.html

 

[daftcombo]

On the foobar2000-resampled file, I put a low shelf brickwall at 14 kHz and listened with headphones (AKG K701). Even at full volume I didn't hear anything.
Afterwards, I normalized the sound to -20 dB. Then I heard a high-pitched noise while slowly increasing volume from 0.

So I guess even the noise-shaping resampler is good enough for playback, the awful looking consequences being inaudible in normal conditions.

 

[bennetng]

On the foobar2000-resampled file, I put a low shelf brickwall at 14 kHz and listened with headphones (AKG K701). Even at full volume I didn't hear anything.
Afterwards, I normalized the sound to -20 dB. Then I heard a high-pitched noise while slowly increasing volume from 0.

So I guess even the noise-shaping resampler is good enough for playback, the awful looking consequences being inaudible in normal conditions.

If I read your screenshot correctly...

Noise shaping is just tradeoff. You get less noise below 15kHz.

 

[daftcombo]

If I read your screenshot correctly...
Noise shaping is just tradeoff. You get less noise below 15kHz.

Absolutely. Either you accept a bit more noise at all frequencies, or more noise in the high frequencies.