Author Topic: Technical Support  (Read 6766 times)

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Technical Support
« on: May 26, 2015, 02:18:27 PM »
We like to encourage the XiSRC community to discuss any technical questions or issues within this category.

Offline johan_van_ooijen

  • Newbie
  • *
  • Posts: 1
Re: Technical Support
« Reply #1 on: September 15, 2015, 01:13:11 PM »
Dear Support,

When I process files for conversion all the meta data is gone. Of course there is meta data in the filename but when I open a file in a meta data editor or Itunes for example all the fields are empty  :-[.

Is there a way to correct this or is this something for a update perhaps?

Thanks in advance.

Best regards,
Johan
« Last Edit: September 15, 2015, 01:45:56 PM by johan_van_ooijen »

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Re: Technical Support
« Reply #2 on: September 16, 2015, 01:25:55 PM »
Dear Johan,

The current version keeps the metadata if you convert from AIFF to AIFF or from FLAC to FLAC.
We are working on metadata support in case of cross conversion between the formats.

Thanks

Stephan
« Last Edit: September 16, 2015, 01:32:28 PM by Xi_Stephan »

Offline Kempy

  • Newbie
  • *
  • Posts: 20
Re: Technical Support
« Reply #3 on: October 19, 2015, 06:59:11 AM »
Hi,

a few comments to my tests with XiSRC v1.0.11, performend on a Sandy Bridge Notebook, i7, 4 cores with HT.

1.)
Async SRC speed, e.g. going from 24bit/96kHz to 16/44.1 is unbelievable slow, sync SRC speed 24/96 to 16/48 looks ok.

2.)
The downsampled file are still contains ISP's, less in quantity but higher in amplitude.

3.)
It is impossible to set the output filename to [Name] only, even when the output directory is different from the source, the start button will not be active.
I have to add something to the resampled filename, meaning I have to re-edit the filenames after resampling.

Kind regards

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Re: Technical Support
« Reply #4 on: October 21, 2015, 08:36:46 PM »
Hi Kempy,

Thanks for bringing this up.

1.) Doing Sample Rate conversion with the highest quality between different sample bases (44.1kHz <-> 48kHz) is the most tricky part.
If we want to downsample from 96 kHz to 44.1 kHz then we must first apply an upsampling of a factor 147 to do afterwards a downsampling by a factor of 320 (96kHz x 147/320 = 44.1kHz).
Any other method tries to do interpolation between samples which creates a higher noise floor caused by the not perfect calculated of new samples.
We apply fast algorithms but the above described approach needs its computation time.
Nevertheless, let us check whether there is room for improvement.

2.) If you downsample a file that contains Inter Sample Peaks then those are still there after down sampling.
XiSRC removes ISPs appearing during up-sampling but to find ISPs before doing downsampling it would be necessary to upsample the file first.
Can you help us in describing the steps you used to create the result of having less ISPs with higher amplitude after downsampling?

3.) Thanks for identifying the issue. We have to check and fix it.
« Last Edit: October 21, 2015, 08:42:21 PM by Xi_Stephan »

Offline Kempy

  • Newbie
  • *
  • Posts: 20
Re: Technical Support
« Reply #5 on: October 22, 2015, 07:00:42 AM »
Hi Kempy,

Thanks for bringing this up.

1.) Doing Sample Rate conversion with the highest quality between different sample bases (44.1kHz <-> 48kHz) is the most tricky part.
If we want to downsample from 96 kHz to 44.1 kHz then we must first apply an upsampling of a factor 147 to do afterwards a downsampling by a factor of 320 (96kHz x 147/320 = 44.1kHz).
Any other method tries to do interpolation between samples which creates a higher noise floor caused by the not perfect calculated of new samples.
We apply fast algorithms but the above described approach needs its computation time.
Nevertheless, let us check whether there is room for improvement.

2.) If you downsample a file that contains Inter Sample Peaks then those are still there after down sampling.
XiSRC removes ISPs appearing during up-sampling but to find ISPs before doing downsampling it would be necessary to upsample the file first.
Can you help us in describing the steps you used to create the result of having less ISPs with higher amplitude after downsampling?

3.) Thanks for identifying the issue. We have to check and fix it.

Stephan,


2.)
sorry,
perhaps I misunderstood the XiSRC feature >Peak Normalisierung, wenn die Quelle „Inter Sample Peaks“ enthält<, thought, the handling is close to AudioRepair.
It's well known that downsampling may introduce slightly higher levels in both sample levels and true-peak.
As you are using linear-phase filters it's not realy an big issue. I found a peak increase when going async from 24bit/96kHz to 16/44.1 in a range of 0.2-0.3dB only and at a different timecode compared to the original.
Min-phase filters would cause much higher level increases.

Perhaps you should add an 'conversion gain' feature to XiSRC allowing negativ values and not related to peak levels (like normalization), simple a volume attenuation by xdB's.

3.)
AudioRepair is affected by that issue too.

Best regards
« Last Edit: October 22, 2015, 09:33:27 AM by Kempy »

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Re: Technical Support
« Reply #6 on: October 22, 2015, 10:05:21 PM »
Hi Kempy,

The issue regarding the target file name has been fixed with today's new release of AudioRepair, AMTRA, HPEX and XiSRC.

Thanks for your help,

Stephan

Offline Kempy

  • Newbie
  • *
  • Posts: 20
Re: Technical Support
« Reply #7 on: November 12, 2015, 08:00:00 PM »
Hi Sirs,

after a few more tests with XiSRC 1.0.12 I found, when downsampling from 24bit-96kHz to 24-44.1, that the integrated loudness/avr. RMS levels of the output-files are attenuated by 1.2-1.3dB, compared to the source-files, when processed with XiSRC.
IMHO a strange and unexpected behave, resampling should not change integrated loudness/avr. RMS levels or only very very slightly.

I checked against with other src's, doing the same downsampling, and none of them alters integrated loudness/avr. RMS levels.

Second thing, why is the dither checkbox unavailable when staying within the same bit-depth ?, considering that the internal audio-path is 64bit float.

Kind regards
« Last Edit: November 12, 2015, 08:03:50 PM by Kempy »

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Re: Technical Support
« Reply #8 on: November 17, 2015, 02:08:55 PM »
Hi Kempy,

Thanks for your intense testing and sorry for my late answer.

Would you be so kind to provide us with your test file used for down sampling so that we can check what happens there?

XiSRC uses flat dithering to truncate from the internal 64Bit to 32Bit/24Bit. Because it makes no sense to disable that dithering the checkbox is not available for the cases of 32Bit -> 32Bit and 32Bit -> 24Bit and 24Bit to 24Bit.
But you discovered a bug in the case of 16Bit to 16Bit conversion, because there it is necessary to select a dithering method. Only if you select 16Bit output it is possible to choose the dithering method.
That will be fixed within the next version.


Offline Kempy

  • Newbie
  • *
  • Posts: 20
Re: Technical Support
« Reply #9 on: November 17, 2015, 06:39:18 PM »
Hi Kempy,

Thanks for your intense testing and sorry for my late answer.

Would you be so kind to provide us with your test file used for down sampling so that we can check what happens there?

XiSRC uses flat dithering to truncate from the internal 64Bit to 32Bit/24Bit. Because it makes no sense to disable that dithering the checkbox is not available for the cases of 32Bit -> 32Bit and 32Bit -> 24Bit and 24Bit to 24Bit.
But you discovered a bug in the case of 16Bit to 16Bit conversion, because there it is necessary to select a dithering method. Only if you select 16Bit output it is possible to choose the dithering method.
That will be fixed within the next version.

Hi Stephan,

thanks for your clarifications.

I was not aware that XiSRC automaticly applies dither internally when going from the internal 64bit float audio-path to 24bit int or 32bit int, so in that case  it make sense that the dither option is disabled.

Sorry, I can't provide my test files.
I simply picked up the first two tracks from Rush's Snakes & Arrows album which I bought from HiresAudio as flac at 24bit-96kHz and downsampled them to 24bit-44.1kHz with XiSRC and then checked for sample-peak and true-peak level changes.
By that, I found that the integrated loudness/avr. RMS levels were attenuated by XiSRC, and, as futher information, in both XiSRC processed files the left channel, which has the lower avr. RMS level (meaning it is the louder channel), always hits 0.00dbFS sample peak.
That might be by fortune or caused by this >Peak Normalisierung, wenn die Quelle „Inter Sample Peaks“ enthält< feature I don't like, thought it is only active when upsampling. It should be optional.
I attached MusicScope's report for the source- and the XiSRC downsampled files.

Btw, is it possible to have a 32bit floating point bit-depth output option in XiSRC ?.
32bit integer is a rarely used format and for further processing 32bit float is more common.

With regards
« Last Edit: November 19, 2015, 09:20:27 AM by Kempy »

Offline geozem

  • Newbie
  • *
  • Posts: 36
Re: Technical Support
« Reply #10 on: November 18, 2015, 11:55:47 AM »
Dear Stephan,

as you are planning to do some updates.

Would you please consider for all target configuration:

Enable/disable switch individually for:
  • inter sample peaks correction and
    frequency respond filtering

I am still not using XiSRC for 24bit 192-96kHz because of the frequency respond filtering standard filtering.

I understand the theory but I hear the negative effect on the sound compared with other tools.

Best regards
geozem

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Re: Technical Support
« Reply #11 on: November 19, 2015, 01:15:34 PM »
Hi Geozem,

We're changing the AudioRepair tool so that it is possible to select the frequency response correction alone without having ISP activated.

As described in detail it is not possible to change the anti-aliasing filter without creating aliasing.
There are a couple of sample rate converters (on the fly or offline) that accept a dedicated amount of aliasing. The high frequency folded back into the audible spectrum creates a perceived change in the tonal balance also used by audio engineers during equalizing to create the illusion of a better sounding record.
So, the artifacts caused bei the SRC-Process are perceived as a better record, but in fact they are definitely not the real thing.

We apply perfectly correct anti-aliasing filtering to suppress any aliasing by -200dB.
Therefore, we achieve a perfect representation of the original record, but compared to other flawed implementations it may sound different.

Best regards,

Stephan

Offline geozem

  • Newbie
  • *
  • Posts: 36
Re: Technical Support
« Reply #12 on: November 20, 2015, 08:15:24 AM »
Dear Stephan,

sorry I think I made my wish for modifications to the wrong tool channel.

Please do NOT chage AudioRepair -> This is perfect as it is. I love it.

Please CHANGE XiSRC with the requested options for individual selection -> Please change only XiSRC. This needs to be changed.

Best regards
geozem


Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Re: Technical Support
« Reply #13 on: November 20, 2015, 11:13:45 AM »
The following comment was regarding XiSRC:

As described in detail it is not possible to change the anti-aliasing filter without creating aliasing.
There are a couple of sample rate converters (on the fly or offline) that accept a dedicated amount of aliasing. The high frequency folded back into the audible spectrum creates a perceived change in the tonal balance also used by audio engineers during equalizing to create the illusion of a better sounding record.
So, the artifacts caused bei the SRC-Process are perceived as a better record, but in fact they are definitely not the real thing.

We apply perfectly correct anti-aliasing filtering to suppress any aliasing by -200dB.
Therefore, we achieve a perfect representation of the original record, but compared to other flawed implementations it may sound different.

Offline Kempy

  • Newbie
  • *
  • Posts: 20
Re: Technical Support
« Reply #14 on: November 21, 2015, 06:45:36 AM »
Hi Sirs,

great to see DSD handling in XiSRC v1.1.0

I did a few fast tests and found

1.)
All kind of DSD (DSD64 and DSD128) files are reported as 352,8kHz/24bit in the input file window, I think that's a bug.
DSD files should be reported as DSDxxx or 1bit-xxxxxMHz.

2.)
Conversion from DSD128 to PCM 24bit-176.4kHz as FLAC output went fine, fast, avr. RMS/Int. Loudness levels are unaltered, but, even the source .dsf file is correctly taged all tags got lost in the process. The resulting FLAC output is without any tag.

3.)
Just for fun, I put a DXD file (352,8kHz/24bit, FLAC, but I don't know if this is a 'real' DXD file, downloaded demo from Norways 2L label) into XiSRC and it was able to convert to PCM 24bit-176.4kHz as FLAC too, and all tags were preserved. OK, finally it is a PCM@FLAC to PCM@FLAC downsampling.

Kind regards
« Last Edit: November 21, 2015, 07:59:49 AM by Kempy »