Skip to main content

Topic: Help me bring clearness to the jungle of resamplers (Read 2513 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • pmattke93
  • [*]
Help me bring clearness to the jungle of resamplers
Hi all,

I don't know if I'm searching on the wrong places...

I have some 48kHz FLACs in my collection, but for compatibility reasons I want to convert them to 44,1kHz - I am thinking about making this conversion final and not keep the original 48kHz since I don't think I would hear any difference.

Still, I would like to know which resampler is theoretically the one with the best performance in foobar:

There is dbPoweramp/SSRC, and there is PPHS which can be selected in the converter setup. PPHS has an ultra mode, I guess this is something like "better accuracy while sacrificing encoding speed". Is that true?

I just want to get the most accurate result possible, I don't care about encoding speed since I don't use the resampler for live listening but only for a one time conversion of my few 48k files.

Which one would you recommend here?

While searching through the web, I can't find anything helpful that encourages my decision. Any tips from the community?

Thanks in advance.

  • DVDdoug
  • [*][*][*][*][*]
Re: Help me bring clearness to the jungle of resamplers
Reply #1
I wouldn't worry too much about it...  I've never heard any difference when converting in either direction between 48kHz (for video) and 44.1kHz (for CD), no matter what resampling software I happen to be using.

...And I ONLY resample when I need 48kHz for video, or 44.1kHz for CD.  If I'm making audio files (MP3s, etc.) from video I leave them at 48kHz.  

I've seen graphs that show a difference, but I'm pretty sure any competent DSP programmer can write a resampler that's better than human hearing, as long as you are not downsampling down to the point where you loose high (audio/audible) frequencies.

  • pmattke93
  • [*]
Re: Help me bring clearness to the jungle of resamplers
Reply #2
I also compared the three resampling algorithms available in stock foobar:

 * dBPoweramp/SSRC
 * PPHS
 * PPHS + Ultra Mode

Converted a 48k file to 44,1k and then back to 48k, then I compared the results and the original in a DAW, twisted the phase of the original and played both together to see what's left:

For PPHS (with and without Ultra Mode) it was mainly noise at about -100dB (at the loudest parts of my test file) and below. With some more high frequency diff at I think about 18-20kHz, where the delta was going up to -80dB. But still of course. Nothing to fear here :D

For SSRC, I was at first surprised that I still could actually quite good listen to the music in the diff between original and twice resampled: Playing both alternating, there was no audible difference for me, but playing the diff sounded a bit like I would grab an EQ and cut off some bass and low mids. I guess there were just some phases different from the original. As far as I know, resampling algorithms often (or always??) are implemented by applying a low pass filter to a zero-stuffed signal. So phase could of course change here, depending on which low pass filter was used. Just some thoughts of mine, don't know if they are true :D can somebody verify?

I know now that both alternatives offered by foobar seem to be transparent to me. But still I want to make a decision. SSRC or PPHS? It's of course not a topic about audible problems. Much more about myself feeling better if I know that I used the resampling method with the best accuracy.

After my tests described above, I would currently go with PPHS + ultra mode, because the noise resulting from the diff was still a bit lower than with PPHS without ultra mode. But due to the phase changes described above for SSRC, I can't determine its quality and there were places on the web were I remember I have read that SSRC would be better than PPHS (but I think the posts / articles where are read that are quite old)

  • lvqcl
  • [*][*][*][*][*]
  • Developer
Re: Help me bring clearness to the jungle of resamplers
Reply #3
There is also SoX resampler plugin (download foo_dsp_resampler_0.8.3.zip from https://hydrogenaud.io/index.php/topic,67373.0.html )

  • Moni
  • [*][*]
Re: Help me bring clearness to the jungle of resamplers
Reply #4
Many, many comparisons: http://src.infinitewave.ca

  • greynol
  • [*][*][*][*][*]
  • Global Moderator
Re: Help me bring clearness to the jungle of resamplers
Reply #5
None of which include double-blind test results of audibility, which, despite the OP's request, ARE ALL THAT MATTER!!!
Is 24-bit/192kHz good enough for your lo-fi vinyl, or do you need 32/384?

  • Moni
  • [*][*]
Re: Help me bring clearness to the jungle of resamplers
Reply #6
There is still value in understanding how they work. If one reads the FAQ or Help sections of that site they will get some context. There is no "ranking." Just objective test data for those curious.

To the OP: SoX or SSRC are just fine. I wouldn't worry about it too much. Like you said, -90 or -100dB is not going to bother you in the slightest. ABX if you wish, but I say just pick one and be done with it.

  • Wombat
  • [*][*][*][*][*]
Re: Help me bring clearness to the jungle of resamplers
Reply #7
For SSRC, I was at first surprised that I still could actually quite good listen to the music in the diff between original and twice resampled: Playing both alternating, there was no audible difference for me, but playing the diff sounded a bit like I would grab an EQ and cut off some bass and low mids...
With SSRC you have the samples slightly shifted in the resulting file. A simple delta does not compensate for this. You may use diffmaker to do a real delta or use SoX standalone that i know compensates this shift.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

  • pmattke93
  • [*]
Re: Help me bring clearness to the jungle of resamplers
Reply #8
What I still don't understand: SSRC and SoX seem to be somehow related. What exactly is the difference between them? Or are these just different names for the same resampling algorithm?

  • bandpass
  • [*][*][*][*]
Re: Help me bring clearness to the jungle of resamplers
Reply #9
No relation; they are just two of the more well-known free resamplers available.

As Wombat mentioned, SSRC (unlike SoX) introduces sub-sample delays, which complicates 'null' testing.

Note that SRC (a.k.a. libsamplerate) is a different resampler again.