Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Does resampling algorithm can influence mp3 decoding quality? (Read 3905 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Does resampling algorithm can influence mp3 decoding quality?

Hi everybody,

I have a friend, the founder of a recently created firm releasing high-end audio files, CDs and vinyles made by digitalising old and forgotten magnetic tapes. His business is becoming to have some really good reputation. But that's not the point.

I wanted to test his ears, and took a 20s excerpt from one of his released HD audio file (which is 24-bit/176.4 kHz mono) and created 5 different flavors:
  • 1 - original : 24-bit / 197.4 kHz
  • 2 - CD 16-bit / 44.1 kHz (avec dither)
  • 3 - mp3 LAME CBR 320
  • 4 - mp3 LAME VBR q0 (254 kb/s)
  • 5 - mp3 LAME VBR q4 (145 kb/s)
And then resampled all this in random order in a 24-bit/176.4 kHz mono file. He listened to this and to my astonishment he was able to tell me the exact quality order (with only a doubt between 3 and 4). As far as I am concerned, I can't distinguish any of those formats :-D Though, I must add that he listened to this on his high-end system, and he deeply knows the source file, since he released it. Really impressive, nevertheless. But that's still not the point.

Since his business is high-end audio, he barely knows mp3 specifications. He claimed that my mp3 compression resampled to 24/176.4 was probably better than the mp3 itself, due to my software resampling algorithm. I believe this is not true, or absolutely negligible, in regard to what is actually lost with the mp3 conversion (and never recovered whatsoever). But I'd like your thoughts on that point!

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #1 –
Quote from: KillBlood on
He claimed that my mp3 compression resampled to 24/176.4 was probably better than the mp3 itself, due to my software resampling algorithm.

Sounds like he doesn't understand what resampling does.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #2 –
Can you elaborate on this?
He's not an engineer, he's a musician (with golden ears, as it seems!). He probably has an idea of what resampling does between different sample rates on uncompressed files, but probably not on an mp3. He told me about high-end machines for doing this: "DCS, MSB, Meitner, Lynx". But obviously nothing related to mp3.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #3 –
Quote from: KillBlood on
claimed that my mp3 compression resampled to 24/176.4 was probably better than the mp3 itself, due to my software resampling algorithm
To be clear, that was not what he was test listening to?

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #4 –
Quote from: Porcus on
To be clear, that was not what he was test listening to?
No. But before trying to answer, he said this (probably because the test seemed a bit harder than what he initially though it would be). But I'm really not sure my Reaper algorithm decoding the mp3 to 24/176.4 does something very different than what does any mp3 player software in realtime.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #5 –
Firstly, make sure Reaper is set to use a high quality resampler. If one is able to ABX anything above 64pt I would consider that person a god, or cheater.
X

Secondly, use a proper ABX tool like the one from foobar instead of simply playing files in random order. Some "hi-end" DACs may not use proper interpolation and can result in audible difference with different sample rates. One may also notice some delay or audible switching noise when the DAC changes sample rates. The foobar ABX tool can avoid this issue by always resample the low sample rate file to match the high sample rate one.

foobar's built-in resampler, RetroArch, has quality settings as well, use at least "Higher" for reliable tests.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #7 –
Both last posts still don't answer to my question, which I will reformulate:
Does the method of decoding/resampling (at hi-res) an mp3 in a DAW can improve the quality in a meaningful way, compared to a naive mp3 player? One more time, my goal was NOT to improve that quality, but rather to have a standard mp3 experience.
That was just something that claimed my friend, which I believe is untrue, in the fear of failing to my test (but he eventually succeeded!).

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #8 –
Decoding:
A DAW with a floating point mp3 decoder can avoid clipping with some mp3 materials.
https://izotope-rx.livejournal.com/5760.html

Resampling:
The mp3 decoder only decodes the mp3 file, resampling is a separate process and can be done with for example, wav, flac and mp3. A reasonable good resampler is not going to degrade the audio quality in an audible way, for example, when downsampling 176.4kHz to 44.1kHz, or resample a decoded mp3 stream to 176.4kHz. Reaper's resampling quality below 64pt sinc in my screenshot for example, can be potentially non-transparent, which means these low quality settings may degrade audio quality in an audible way when resampling mp3, wav, flac and so on.

So the answer is:
[1] A poor enough resampler may degarde decoded mp3 files in audible ways, so your friend could be right depends on the resampler's quality settings.

[2] A high quality mp3 decoder is not going to clip the waveform while decoding, a "naive mp3 player" on the other hand may, and the differences are potentially audible depends on listeners and audio materials.

[3] Read my previous post again and do what I mentioned (e.g. foobar ABX) to test your friend again if your have another chance, as your test methodology is fragile.


Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #9 –
Of course he can hear SEE the difference, if you know what I mean...
I am pretty sure that any experienced user here can provide you with valid ABX logs in like 2 minutes.

Quote from: KillBlood on
founder of a recently created firm releasing high-end audio files

Your friend's job is selling ultra mega super high quality audio files. Of course he will tell you that audible difference is obvious.
But guess what, you don't have ultra mega super high end system like he does, so you cannot tell the difference...

To answer your question, with modern resampling and dithering there will be no audible differences.
16/44 is same as 32/384.  Later only takes more space, that's it.
gold plated toslink fan

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #10 –
I know what are ABX tests but my test was rather a ranking test, not an identification test. I don't see the flaws in my test, since I provided him 5 files reencoded in the same format (24/176.4). I'm not aware of any software capable of analysing this file and saying it has been encoded in mp3 CBR 320 before being decoded/resampled. And I don't think he would have lied to me, whatsoever.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #11 –
So did you actually watching him doing the listening tests, or just sent the files to him without knowing what he did later on, and just got a reply from him?
Quote from: KillBlood on
I wanted to test his ears, and took a 20s excerpt from one of his released HD audio file
So 20 seconds for each file, perfectly fine with TOS9.
You can upload the files without telling which is which, upload the files as FLAC and let us guess the order.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #12 –
Ok, here's the files! https://we.tl/t-vwvo5uzd21?src=dnl
I provided 5 separate files, and a unique file with all in the same order.
Sorry for the size, and I could have encoded in mono since the source is mono (for some reasons, he releases mono recordings as stereo files, and didn't want to say my why - a secret!).
Also, in the info file, the 176.4 is wrong.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #13 –
1.aif - CD 16-bit / 44.1 kHz (avec dither)
2.aif - original : 24-bit / 176.4 kHz
3.aif - mp3 LAME VBR q4 (145 kb/s)
4.aif - mp3 LAME VBR q0 (254 kb/s)
5.aif - mp3 LAME CBR 320

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #14 –
Very good, you just inverted two consecutive items (like my friend, but not the same ones).
How did you do?

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #15 –
1 - 320
2 - Original
3 - V4
4 - V0
5 - CD

Large image in spoiler.
Spoiler (click to show/hide)

- 320 MP3s will have sharp cut-off at around 20 kHz, number 1.
- MP3 is limited to 48 kHz, CD to 44 kHz, so anything above that has to be original file, in this case number 2.
- V4 MP3s will have very obvious cut-off at 16 kHz, so number 3.
- V0 MP3s will have lowpass disabled by default, you can see 16 kHz cut-off but you can also see that it hits 24 kHz, number 4.
- Nice smooth cut at 22 kHz, CD is number 5.
gold plated toslink fan

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #16 –
Quote from: Markuza97
1 - 320
2 - Original
3 - V4
4 - V0
5 - CD
That's it. Nice! Indeed, the spectrum is very clear. I do believe, however, that my friend was able to make that ranking by ear, does it seem impossible to you?
He insisted to redo the test in front of me, on a longer portion, and stereo, which I will probably do next time I meet him.

However, you still haven't exactly answer to my specific question, about the potential enhancement of resampling after decoding mp3. But the massive loss of high frequency, as you've shown, is definitive, and no algorithm is able to restitute them.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #17 –
Sigh. I should have spent more time to inspect the spectrum more carefully instead of doing some quick looking  :-[
But yes, this kind of cheating is possible.
 

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #18 –
Quote from: KillBlood on
He claimed that my mp3 compression resampled to 24/176.4 was probably better than the mp3 itself, due to my software resampling algorithm. I believe this is not true, or absolutely negligible, in regard to what is actually lost with the mp3 conversion (and never recovered whatsoever). But I'd like your thoughts on that point!
To directly answer your original question, no. A resampler doesn't have the ability to reconstruct information lost during a lossy encoding process. It only changes the sampling rate of a PCM signal while adding acceptable amount of distortion. Someone might notice the difference caused by a crappy resampler, like it, and claim that as an improvement. However, increase in preference doesn't necessarily mean improvement.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #19 –
Quote from: OLPP on
no. A resampler doesn't have the ability to reconstruct information lost during a lossy encoding process. It only changes the sampling rate of a PCM signal while adding acceptable amount of distortion.
Just, be a bit careful about what could happen in certain circumstances:
Source is a floating-point format, volume could exceed digital full scale. It is not given that the playback chain protects against clipping.
Target format is integer (this case, 24-bit integer).
It would not be insane if DSP software that performs a conversion from float to integer - with or without resampling! - would take measures to protect against clipping, possibly lowering volume.
If the playback chain does not offer the same protection, the source may clip while target will not.

If they are compared without volume adjustment, humans are prone to preferring a slightly louder signal, everything else equal. (That of course makes listening tests without volume matching inadequate for the usual purpose - which is a different side to the same story though.)

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #20 –
Quote from: KillBlood on
He told me about high-end machines for doing this: "DCS, MSB, Meitner, Lynx". But obviously nothing related to mp3.
Quote from: bennetng on
Some "hi-end" DACs may not use proper interpolation and can result in audible difference with different sample rates.
Here is an example:
https://www.stereophile.com/content/meitner-ma3-integrated-da-processor-measurements
Meitner, Fig.2

You can see the Meitner DAC above does different things with white noise vs tone input. The tone input is properly bandlimited while the white noise is not. Typical DACs usually look like this:
https://www.stereophile.com/content/okto-dac8-stereo-da-processor-measurements
Okto, Fig.6, AFRLP filter


Subsequently, when this Meitner DAC sees two tones in the input, the measurement result is no longer properly bandlimited:
Meitner, Fig.12


Yet the Okto always bandlimit the input, as long as the same filter is being used (AFRLP, which I suppose stands for Apodizing Fast Roll-off Linear Phase, one of the built-in filters from ESS chips)
Okto, Fig.17


All the tests above were using 44.1k input, and I don't know what the Meitner would do when using 176.4k input.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #21 –
Now you have all the answers you were looking for :)

One more thing I would like to mention...
See that straigh line at around 15.5 kHz? Some equipment is not shielded properly.
In this specific case, it is in audible range, but it is masked by music and noise so you are not going to hear it.
Still worth checking out.
gold plated toslink fan

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #22 –
Yes, thanks, I will read the last posts carefully.
I was wandering about this line. Do you mean it comes from a lack of shielding in the 1969 equipment, or in the digitalisation process?

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #23 –
Horizontal scanning frequency of a CRT monitor most probably.

Re: Does resampling algorithm can influence mp3 decoding quality?

Reply #24 –
Quote from: KillBlood on

That's it. Nice! Indeed, the spectrum is very clear. I do believe, however, that my friend was able to make that ranking by ear, does it seem impossible to you?
He insisted to redo the test in front of me, on a longer portion, and stereo, which I will probably do next time I meet him.

You've got to be the most gullible person ever. Your friend is running a business, based on audiophile woo. Of course he ca never admit (or subject himself to a fair test) that there's no perceptible difference between CD audio and his files with >22khz imperceptible noise.