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: SSRC 1.31 (Read 25016 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Re: SSRC 1.31

Reply #75 –
Quote from: greynol on
Are there recommendations for how testing should be done?

SSRC uses two filters made by Kaiser window.
I am going to add options for setting parameters for those Kaiser windows.
I briefly searched the Internet for good explanation for Kaiser window, but I haven't found.
I will write tutorial for it.

Quote from: Wombat on
Trying to convert 24/96 to 16/44.1
pdf type 3 which i guess tries to avoid dither on silence crashes with message:
File: ssrc.c, Line 1845
Expression: shaper != NULL

profile long crashes with a manual set pdf type
profile fast is not recognized

Thank you for testing.
Profile fast is renamed to short. Anyway I am going to create new profiles according to your feedback.
PDF type 3 does not work, but anyway that is a very experimental option, and I see some problems with it.

Re: SSRC 1.31

Reply #76 –
Quote from: Naoki Shibata on

Quote from: Wombat on
Trying to convert 24/96 to 16/44.1
pdf type 3 which i guess tries to avoid dither on silence crashes with message:
File: ssrc.c, Line 1845
Expression: shaper != NULL

profile long crashes with a manual set pdf type
profile fast is not recognized

Thank you for testing.
Profile fast is renamed to short. Anyway I am going to create new profiles according to your feedback.
PDF type 3 does not work, but anyway that is a very experimental option, and I see some problems with it.


I confirm that error with pdf 3, however generally I use TPDF (pdf 2).

Re: SSRC 1.31

Reply #77 –
Regarding testing, I'm talking about transparency vs audible artifacts and how it relates to real-world listening to real-world content.

Otherwise, I don't see how any of this actually matters.  Please enlighten me if you can.

Re: SSRC 1.31

Reply #78 –
I think we've been here before.  Dithering isn't important, but hey, may as well do it (per Monty's video).

About the only thing bordering on a practical use-case for dithering is listening to fade-outs at elevated volume.

TPDF is the mathematically correct way to dither and should be chosen unless there is a real reason for using something more exotic.

Placebophiles should note that using shaped dither worsens time resolution.

Re: SSRC 1.31

Reply #79 –
Quote from: greynol on
Regarding testing, I'm talking about transparency vs audible artifacts and how it relates to real-world listening to real-world content.

Otherwise, I don't see how any of this actually matters.  Please enlighten me if you can.
I admit thats the real problem. Like mentioned before my X-Fi driving HD-590 is loud enough to listen everything comfortable. 16bit flat dither is already unaudible when not amplified. Any classic lover with a unreal silent recording to share with us?
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: SSRC 1.31

Reply #80 –
While I don't want to be the one to interrupt you guys from jacking off all over yourselves, just be honest about what you're doing.

Naoki is asking for preferences and/or testing and I was afraid he wasn't going to get meaningful input, considering the history of one of the current participants in the discussion.

That you've also crossed the line into placebophilia on this topic is something I may never understand.

Anyway, I think I'm done thread-crapping. Carry on.

Re: SSRC 1.31

Reply #81 –
Okay. I see. Anyway, I will add the options for setting filter parameters.
That is not hard.
I will decide if the feedback is meaningful.

Re: SSRC 1.31

Reply #82 –
Quote from: greynol on
That you've also crossed the line into placebophilia on this topic is something I may never understand.
Me must have got confused when lately someone asked why foobar uses that strong shaped dither and i started again this dither thing. Some years back there was also this "dither shootout" test online and its construct worked well to show how it can work.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: SSRC 1.31

Reply #83 –
Quote from: Naoki Shibata on
Okay. I see. Anyway, I will add the options for setting filter parameters.
That is not hard.
I will decide if the feedback is meaningful.

Thank you for developing SSRC further. If there are new options I will try (at least some of) them. Also I have worked on very simple batch script that utilizes FLAC binary to convert FLAC to WAV, then do conversion with SSRC and convert (losslessly compress) the result back to FLAC. I am probably unable to do testing according to strict methodology that is usual here, but I will try to utilize the new version and let you know what my experience is and you as author can of course decide about the feedback.

Re: SSRC 1.31

Reply #84 –
Just an additional information, I tried tonight SSRC 1.32 with parameters lower

ssrc_hp.exe --rate 44100 --bits 16 --dither 2

on a number of my favourite tracks and the results are very good, both digitally (no issues with conversion) and sonically (no issues with listening to them). I also made a CD from those tracks and even on my older CD player they seem to be OK.

Re: SSRC 1.31

Reply #85 –
Just an additional information, I tried tonight SSRC 1.32 with simple parameters lower

ssrc_hp.exe --rate 44100 --bits 16 --dither 2

on a number of my favourite tracks and the results are very good, both digitally (no issues with conversion) and sonically (no issues with listening to them - as far as I can tell). I also made a CD from those tracks and even on my older CD player they seem to be OK.

Re: SSRC 1.31

Reply #86 –
I did some additional testing and have to say that current version of SSRC (1.32) works very well. I did even some ABX tests against SoX with standard settings (95 linear VHQ), unfortunately the results are not statistically significant (e.g. 9/16), but  generally  I can say - sorry for subjective statement - that SSRC conversion sounds very good, especially on some tracks that are  slightly "softer" than SoX's ones. Overall I enjoy SSRC results. Considering the filter length I did not observe audible effect with long so I use the default quality. I use SSRC together with flac.exe and metaflac.exe to work with losslessly compressed audio.

So if the next version is released with some changes, please keep the current resampler settings the default and/or selectable the same way you did with dithering (keeping ID 90-92 available in addition to new curve). If a tutorial is written the effects of particular settings could be described deeper and the recommended defaults given.

Thank you for developing the great resampler!

Re: SSRC 1.31

Reply #87 –
Quote from: jumpingjackflash5 on
I did some additional testing and have to say that current version of SSRC (1.32) works very well. I did even some ABX tests against SoX with standard settings (95 linear VHQ), unfortunately the results are not statistically significant (e.g. 9/16), but  generally  I can say - sorry for subjective statement - that SSRC conversion sounds very good, especially on some tracks that are  slightly "softer" than SoX's ones.
SSRC resampler sounds softer then SoX? I seldom did read such nonsense over here :)
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: SSRC 1.31

Reply #88 –
Quote from: Wombat on
Quote from: jumpingjackflash5 on
I did some additional testing and have to say that current version of SSRC (1.32) works very well. I did even some ABX tests against SoX with standard settings (95 linear VHQ), unfortunately the results are not statistically significant (e.g. 9/16), but  generally  I can say - sorry for subjective statement - that SSRC conversion sounds very good, especially on some tracks that are  slightly "softer" than SoX's ones.
SSRC resampler sounds softer then SoX? I seldom did read such nonsense over here :)

It may depend on what we mean under that subjective description - maybe the word is not chosen well.. And the problem is that ABX testing I have done unfortunately does not provide staistically significant results to argue. Anyway I have no problems with SSRC conversions I have done at current settings.

Re: SSRC 1.31

Reply #89 –
Quote from: Wombat on
Quote from: jumpingjackflash5 on
I did some additional testing and have to say that current version of SSRC (1.32) works very well. I did even some ABX tests against SoX with standard settings (95 linear VHQ), unfortunately the results are not statistically significant (e.g. 9/16), but  generally  I can say - sorry for subjective statement - that SSRC conversion sounds very good, especially on some tracks that are  slightly "softer" than SoX's ones.
SSRC resampler sounds softer then SoX? I seldom did read such nonsense over here :)

Thank you for testing SSRC.
I will set rules for ABX testing when the next version will be released.
I would like to know the minimum filter lengths with which people cannot distinguish between the original and converted wav files.

Re: SSRC 1.31

Reply #90 –
Thank you very much, Naoki. I suppose that current default settings will stay available for common conversions and new will be made available first for testing purposes, so that one can utilize SSRC for "trusted" conversions the same way as it can be done now.

Re: SSRC 1.31

Reply #92 –
By trusted (already proven working) I mean current (1.32 and earlier) settings of Ssrc resampler that have already been tested in the field with good results. Since Naoki wants to introduce/test also new options I wanted to assure him that current options work well and provide good sound experience.

Re: SSRC 1.31

Reply #93 –
Quote from: jumpingjackflash5 on
...I wanted to assure him that current options work well and provide good sound experience.
It is a big step forward SSRC has the 'jjf5 trusted' certificate now.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Re: SSRC 1.31

Reply #94 –
Quote from: Wombat on
Quote from: jumpingjackflash5 on
...I wanted to assure him that current options work well and provide good sound experience.
It is a big step forward SSRC has the 'jjf5 trusted' certificate now.

Not only mine, but also more experienced users and also authors of popular sw like foobar who included ssrc routines in their sw as plugins.

Re: SSRC 1.31

Reply #95 –
Quote from: Wombat on
It is a big step forward SSRC has the 'jjf5 trusted' certificate now.
I can't say I didn't already warn Shibata Naoki, though it should be pretty evident by now that jjf5's testimony is unreliable.

Re: SSRC 1.31

Reply #96 –
I've found another unexpected behaviour with ssrc.

Here, a 7200Hz tone sampled both at 32 kHz and at 44.1 kHz is decimated to 16 kHz using ssrc and two other well-known resamplers:
Code: [Select]
sox -r 44100 -n -twavpcm in1.wav synth 10 sin 7200 gain -1
sox -r 32000 -n -twavpcm in2.wav synth 10 sin 7200 gain -1

for r in sox lsr ssrc; do
  for n in 1 2; do
    case $r in
      ssrc) ./ssrc --quiet --rate 16000 in$n.wav out$n.wav ;;
      sox) sox in$n.wav out$n.wav rate 16000 ;;
      lsr) sndfile-resample -to 16000 in$n.wav out$n.wav >/dev/null ;;
    esac
  done
  echo; echo $r
  sox -M out[12].wav -n trim 1 8 stats 2>&1 | grep -E "Left| lev "
done
Which gives (where Left and Right are the results from 44100 & 32000 resp.):
Code: [Select]
sox
             Overall     Left      Right
Pk lev dB      -1.00     -1.00     -1.00
RMS lev dB     -4.01     -4.01     -4.01

lsr
             Overall     Left      Right
Pk lev dB      -3.14     -3.14     -3.14
RMS lev dB     -6.15     -6.15     -6.15

ssrc
             Overall     Left      Right
Pk lev dB      -1.76     -6.91     -1.76
RMS lev dB     -6.53     -9.87     -4.66
Conventional wisdom says that a decimation filter should be set relative to the nyquist frequency of the destination sample-rate.  Here, we have two decimations to the same destination rate and with sox and libsamplerate there is no difference in the results, but with ssrc there is over 5 dB difference between the two; i.e. it seems to vary the decimation filter with the source sample-rate.  Not sure if this is a bug or if there is any other explanation.
 

Re: SSRC 1.31

Reply #97 –
Here's what this looks like using a sweep instead of a tone (x-axis is frequency in kHz):


Re: SSRC 1.31

Reply #98 –
Today I released version 1.33 of SSRC.
Only performance and compatibility enhancements are added in this release.
I am sorry, but I was too busy in May, and I will look into the issue reported by bandpass.

Re: SSRC 1.31

Reply #99 –
Thank you for new release anyway.