Resampler plugin, uses SoX 14.2.0 resampling routines Options

[lvqcl]

Uploaded here

Good quality, fast resampler (~2 times faster than PPHS Ultra, although ~2.5 times slower than regular PPHS). Minimum / intermediate / linear phase.
Any comments?

Added: If you want to know what settings are best:
1. Read SoX FAQ, "What are the best 'rate' settings to resample a file and retain the highest quality?"
2. This post: http://www.hydrogenaudio.org/forums/index....st&p=626176 (an excerpt from SoX help)
3. Feel free to experiment and decide what's best for you.

This post has been edited by lvqcl: Apr 10 2009, 20:37

[patul]

Downloading now & will try soon. I use SSRC however. Do you have any detailed comparison between your SoX based resampler vs PPHS, SSRC??

Thanks

This post has been edited by patul: Nov 19 2008, 06:44

[AFAIK]

well... somebody made this test


http://src.infinitewave.ca/

[Canar]

Now does this resampler use the "High" or "Very High" setting?

Thanks for the component, by the way.

[AFAIK]

I think i found a bug


when using in dsp chain and play a 22050Hz file, the playback time displayed in foobar is slower than it should be
and it sometimes even stall at 0:00

there might be something wrong...


btw. i love this component

[lvqcl]

Now does this resampler use the "High" or "Very High" setting?

Thanks for the component, by the way.

Sorry, I don't fully understand your question.
SoX effect named "rate" has 5 quality presets: quick, low, medium, high, very high.

CODE

    |      Quality   Band-  Rej dB   Typical Use       |
    |                width                             |
    |-q     quick     n/a   ≈30 @    playback on       |
    |                        Fs/4    ancient hardware  |
    |-l      low      80%    100     playback on old   |
    |                                hardware          |
    |-m    medium     95%    100     audio playback    |
    |-h     high      95%    125     16-bit mastering  |
    |                                (use with dither) |
    |-v   very high   95%    175     24-bit mastering  |

Also, 'quick' and 'low' quality levels doesn't allow to override options like aliasing, steep filter and phase response. So I disabled them. I don't think that foobar users play music on ancient or old hardware

[lvqcl]

when using in dsp chain and play a 22050Hz file, the playback time displayed in foobar is slower than it should be
and it sometimes even stall at 0:00

Thanks, I'll investigate this.

[Canar]

Please ignore my earlier comment. I missed the configuration somehow. This looks great!

[vigylant]

well... somebody made this test
http://src.infinitewave.ca/

The SSRC is better than this component, right?
Because if you select Passband/Transition, you can see that the SSRC is more similar to the "ideal filter".

[lvqcl]

The SSRC is better than this component, right?
Because if you select Passband/Transition, you can see that the SSRC is more similar to the "ideal filter".

I'm afraid that it isn't so simple. Look at:

http://www.hydrogenaudio.org/forums/index....ost&p=81050

http://www.hydrogenaudio.org/forums/index....ost&p=29125

But I don't know of the audibility of ringing at Nyquist limit.

I think i found a bug


when using in dsp chain and play a 22050Hz file, the playback time displayed in foobar is slower than it should be
and it sometimes even stall at 0:00

there might be something wrong...


btw. i love this component

Bug confirmed and (I hope ) fixed. Updated version is available. Thanks!

[vigylant]

Hmm, ok, i take it this one is better then
But im wondering: What should i use for samplerate conversion? (From 44100 to 96000)
My soundcard (Onkyo SE-90PCI, highend stereo soundcard, VIA Envy24 chip) or this plugin, any idea which one is better?

Probably hard/impossible to tell without actually testing my soundcard

This post has been edited by vigylant: Nov 19 2008, 23:58

[bandpass]

The SSRC is better than this component, right?
Because if you select Passband/Transition, you can see that the SSRC is more similar to the "ideal filter".

I'm afraid that it isn't so simple. Look at:

http://www.hydrogenaudio.org/forums/index....ost&p=81050

http://www.hydrogenaudio.org/forums/index....ost&p=29125

lvqcl, good work with the port, and good pointers to some well informed posts.

vigylant, the bandwidth of the SoX SRC is configurable up to 99.7% which pretty much matches that of SSRC (standard). But be aware, that the closer you get to the nyquist, the more artifacts you get; look closely at the transition graphs for the Shibatch SRC and you'll see that (esp. for the hp version) the transition lies to the right of the ideal filter's -- i.e. aliasing is occurring.

-bandpass

[jaro1]

I had carefully compared this SRC, based on SOX 14.2.0 with other foobar resamplers and it is sonically the best resampler out there, IMHO of course. On this SRC I appreciate very low artifacts, frequency responce, good transition and very low noise floor.
Then I found this, maybe well known side yet (http://src.infinitewave.ca), which basically confirmed my experience. SOX with VHQ settings is according to this side for me in pair with iZ RX Advanced (High steepness), which is in some parameters maybe very slightly better, but has also a little higher noise floor, which is on sweep test result picture quite visible and therefore maybe for highend headphones or speakers also audible.
So many many thanks to the author for hard work and time for creating and uploading this free soft, which will certainly improve listening experience for people with low/medium quality sound cards/int. chips. To author, I intercede for future development of this great plug-in to make it even better, btw I didn't find any bugs in this version yet.
So thank you very very much again

[Canar]

@jaro1: I am not convinced that you did, indeed, hear a difference. The quantifiable differences between PPHS and this resampler are below the threshold of human hearing. Please re-read our terms of service, paying attention to point 8.

That is not to diminish the efforts of the developer of this component. I am now using it myself for high quality work, due to the objective differences. Subjectively, however, I am not convinced it is possible to discern between PPHS and the SoX resampler. The site you cited primarily concerns itself with inaudible qualities.

[hidn]

question which I must ask. He as well as PPHS Ultra?

[jaro1]

@jaro1: I am not convinced that you did, indeed, hear a difference. The quantifiable differences between PPHS and this resampler are below the threshold of human hearing. Please re-read our terms of service, paying attention to point 8.

That is not to diminish the efforts of the developer of this component. I am now using it myself for high quality work, due to the objective differences. Subjectively, however, I am not convinced it is possible to discern between PPHS and the SoX resampler. The site you cited primarily concerns itself with inaudible qualities.

Because of very short time I could not write all circumstances of listening tests I'done, now I can only add, that I wasn't alone, there were three of us, loudspeakers-B&W Sig. Diamond; headphones - AKG K701.
But according to service terms you are absolutely right, but I wanted because of time shortness write something about our SUBJECTIVE observations and at least say thanks to the author, sorry for forum rules breaking and also for my bad english.

[Jorgo]

This seems to be a very good resampler. From my limited understanding, it rivals the SSRC resampler but uses less CPU.

A visual comparison can be seen at
http://sox.sourceforge.net/SoX/Resampling
http://sox.sourceforge.net/rate-44k1-96k.png

The example of VERY HIGH (-v) looks almost the same as the Secret Rabbit result with -c0 and
Secret Rabbit -c1 looks very close to MEDIUM with 90% bandwidth (-mb 90). Other than that, I don't know enough to be able to interpret the impulse response per se.

Here are some more pointers from the SOX manual:

Quality Band-width Rej dB Typical Use
-q quick n/a ~=30 @ Fs/4 playback on ancient hardware
-l low 80% 100 playback on old hardware
-m medium 95% 100 audio playback
-h high 95% 125 16-bit mastering (use with dither)
-v very high 95% 175 24-bit mastering

This post has been edited by Jorgo: Nov 23 2008, 20:26

[hidn]

someone who can compile and lay out icl version?

[lvqcl]

Version 0.3.0 is out. It seems that resampling routines are not thread-safe, so I changed them a little.

By the way, I think that "Resampler based on SoX 'rate' effect" is a bit awkward name. Any ideas how to name this plugin more elegantly?

[Canar]

SoX Resampler.

Resampler (SoX) in the DSP list.

[.halverhahn]

Thank You for transfering this great resampler to f2k!

[Jorgo]

0.4.0... uh, what? Please don't rename stuff so it gets broken in the DSP setup... the names should be kept between updates.

[lvqcl]

0.4.0... uh, what? Please don't rename stuff so it gets broken in the DSP setup... the names should be kept between updates.

0.4.0 version is in alpha stage

[jaro1]

Hi, I know it's a little offtopic, but I wanted to ask lvqcl to something, because I suppose he knows SOX internal routines quiet good yet.
Would it be possible in the future make a plugin, which implements SOX deemphasis filter? (deemph: ISO 908 CD de-emphasis (shelving) IIR filter)

Till now I use for it wav impuls responses with FB2Ks Convolver plugin, but this isn't a right choice according to this german page:
http://www.radonmaster.de/robernd/tAFILTER.html (thanks to author and Surfi for the link)
On this side there are packed wav signals, with explanation, for testing different deemphasis filters ( there is also filter of old SOX version, which results very badly, it altered also stereo image... )
Filter from recent SOX version is perfect, but solution with Convolver plugin (whatever impulse I use) alters a signal after closser look in Audition quiet a lot from what it should do, although frequencygoing could be very close and therefore after listening undistinguishable.

For me is use of Convolver plugin, at least for this purpose, a little unpractical and ultimately inaccurate, so therefore this idea. By the way, it may or may not be a problem, but I think SOX deemph filter has 32bit internal precision, which is, on the other side, for this kind of operation absolutely necessary.

Once more, sorry for offtopic and thanks for your opinion.

This post has been edited by jaro1: Mar 10 2009, 11:48

[lvqcl]

Hi, I know it's a little offtopic, but I wanted to ask lvqcl to something, because I suppose he knows SOX internal routines quiet good yet.

I think bandpass knows them much more

Filter from recent SOX version is perfect, but solution with Convolver plugin (whatever impulse I use) alters a signal after closser look in Audition quiet a lot from what it should do, although frequencygoing could be very close and therefore after listening undistinguishable.

Seems strange for me. Try to use this impulse: deemph_SoX.wav

For me is use of Convolver plugin, at least for this purpose, a little unpractical and ultimately inaccurate, so therefore this idea. By the way, it may or may not be a problem, but I think SOX deemph filter has 32bit internal precision, which is, on the other side, for this kind of operation absolutely necessary.

I tested SoX and Convolver with my impulse: difference between them is about -130 dB. So convolver is quite accurate.
And don't use "Auto level adjust" option.

This post has been edited by lvqcl: Feb 14 2013, 16:19