Author Topic: Technical Support  (Read 5744 times)

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 458
Technical Support
« on: September 01, 2017, 07:58:23 PM »
We like to encourage the XiFEO community to discuss any technical questions or issues within this category.

Offline scan80269

  • Newbie
  • *
  • Posts: 6
Re: Technical Support
« Reply #1 on: September 04, 2017, 04:44:43 AM »
Thank you for implementing the new filtering options in XiFEO 1.1.0!

With this version, I ran into an issue when selecting the "High Cut" filtering option.  The processing of a track would always stop at 49% or 50%, with the State becoming "Error".  This is apparently at the end of the Pre-Processing phase.  I have tried 24/96K, 24/176.4K, 24/192K and 24/352.8K content with the same result.  This issue does not occur with the other filtering options, including Automatic.

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 458
Re: Technical Support
« Reply #2 on: September 04, 2017, 12:41:47 PM »
Thank you for identifying the issue.
The bug-fix release is planned for today.

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 458
Re: Technical Support
« Reply #3 on: September 04, 2017, 09:47:46 PM »
XiFEO Version 1.1.1 fixes the issue.

https://www.xivero.com/xifeo/

Offline scan80269

  • Newbie
  • *
  • Posts: 6
Re: Technical Support
« Reply #4 on: September 10, 2017, 12:40:43 AM »
With XiFEO 1.1.1 and the Automatic setting for Bit Truncation, some 24/352.8K FLAC files are truncated by 8 bits, and the output file is 16-bit instead of 24-bit.  Is this correct behavior?

For example, the following DXD resolution files from 2L.no website:

http://www.lindberg.no/hires/test/2L-092/2L-092_stereo-DXD_01.flac
http://www.lindberg.no/hires/test/2L-106/2L-106_stereo_DXD-352k_MAGNIFICAT_04.flac
http://www.lindberg.no/hires/test/2L-038_stereo_FLAC_352k_24b_01.flac
http://www.lindberg.no/hires/test/2L-082_stereo-DXD_01.flac

These files all get 8 bits of truncation when processed by XiFEO 1.1.1.  I recall the previous XiFEO 1.0.3 processing the same files resulted in 5 or 6 bits of truncation, so wondering which truncation depth is the correct one.



Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 458
Re: Technical Support
« Reply #5 on: September 10, 2017, 05:53:14 PM »
Yes, it is correct that XiFEO outputs a 16 bit file if it works on a 24 bit input file that contains 8 bits of noise.

Let's dig into more details about your test files.
1.) Those files are DXD which is the intermediate file format to work on DSD recordings.

2.) All files show the high noise floor which is shaped in a way that it increases beyond the critical audio bandwidth (up to -60 dB).

3.) XiFEO applies statistical methods to find the actual noise floor within the audio bandwidth.

4.) To prove that the results are valid we've attached MusicScope measurements of the "2L-092..." audio file which show the noise density of the original and the bit truncated file at a dedicated position within the spectrogram.

The upcomming version of the MusicScope allows precise noise density measurements.

The noise floor beyond the audio content (above the cut-off frequency) has been measured at around -91dB. Well, it is not easy to measure the real noise floor without applying statistical methods, but in this case we’re safe to say that there is no audio content but only noise.

The XiFEO processed 16 bit output file reaches up to -96 dB (plus dithering), so that there is effectively no difference between both files in terms of audio quality.

Only during digital fade-in and fade out, where the levels could reach 0 dBFS, the quantization noise could be higher. Well, we're finally interested in the noise floor of the audio content itself and that is not increased.

Furthermore, XiFEO filters the high frequency part (if automatic filtering is selected), that just contains noise and reaches a noise floor of around -96 dB which is quite a bit better than having the high DSD noise floor of up to -60 dB.

Nevertheless, we don't say that this is a bad recording. Especially during the recording process, it is important to have enough headroom (e.g. 24Bit) to achieve the best SNR possible. Even by having just 16 effective bits we can say that this recording really provides around 96 dB of SNR + dithering.

« Last Edit: September 10, 2017, 06:09:42 PM by Xi_Stephan »