Author Topic: Feature check: Peak Normalisierung, wenn die Quelle „Inter Sample Peaks“ enthält  (Read 2098 times)

Offline geozem

  • Newbie
  • *
  • Posts: 36
As you see I am planing to buy more from you  :) so I am checking

I have bought some 24bit 96khz material which has a level put to 0db as highest for what so ever artistic reasons.

I would like to use the feature "Peak Normalisierung, wenn die Quelle „Inter Sample Peaks“ enthält for transcoding only.
The same what I am using with Audiorepair tool of use for 16 bit material.
There is no tick which I could activate manually.

what I have done:
Alt-J - An Awesome Wave - 03 - Tessellate.flac -> XiSRC -> Alt-J - An Awesome Wave - 03 - Tessellate - [24 - 96.0 kHz - None].flac
result: no automatic activation or usage of  feature "Peak Normalisierung, wenn die Quelle „Inter Sample Peaks“


So far the  feature "Peak Normalisierung, wenn die Quelle „Inter Sample Peaks“ will be only acitvated if I change from example:
Alt-J - An Awesome Wave - 03 - Tessellate.flac -> XiSRC -> Alt-J - An Awesome Wave - 03 - Tessellate - [16 - 44.1 kHz - Flat].flac


Moreover if I compare:
created  by XiSRC: Alt-J - An Awesome Wave - 03 - Tessellate - [16 - 44.1 kHz - Flat].flac
with created by Audiorepair: Alt-J - An Awesome Wave - 03 - Tessellate - [ISPR=ON Steep 44.1kHz].flac
then I got different volumes slightly. The result form Audiorepair is lower.

I have used to transcode from 24bit to 16 bit with dBpoweramp

Can you please check?
« Last Edit: August 26, 2015, 10:13:20 AM by geozem »

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Hello Geozem,

Thanks for using the software.

The approach of AudioRepair and XiSRC to avoid Inter Sample Peaks is slightly different.
AudioRepair oversamples by a factor of 32 (1.4112MHz) to detect ISPs and adapts the amplitude of the original signal accordingly.

XiSRC upsamples the signal to the target sample rate and checks whether there are Inter Sample Peaks detectable, because a higher sample rate makes them visible, and adapts the amplitude of the signal to avoid or reduce the impact of Inter Sample Peaks. That is quite special because not many Sample Rate Converters care about ISPs.

To see the True Peak Level of the outputs from the AudioRepair and XiSRC tool you could use the MusicScope which provides you in both cases with the TPL and therefore a great possibility to compare both results.
There shouldn't be much of a difference.

PS:
If you don't change the sample rate then XiSRC does not suppress ISPs and therefore does not adapt the signal amplitude.

Best regards,

Stephan

Offline geozem

  • Newbie
  • *
  • Posts: 36
Dear Stephan,

thank you for confirming my results as features of the tool.

On the other hand I have started to realize how complex this topic is and how many factors are relevant to reproduce good sound. I need to need to deep dive here.

What I don't get so far why is the 24bit material not so strongly effected by ISPs?

Potentially you have a simple answer. Is this related to the Tiefpassfilter?

Best regards
Geozem
« Last Edit: August 28, 2015, 01:54:46 PM by geozem »

Offline Xi_Stephan

  • Global Moderator
  • Sr. Member
  • *****
  • Posts: 447
Hello Geozem,

Your're absolute right. The whole topic is quite complex and we're learning every day new thinks we like to implement in our software.

The ISP issue becomes less critical with increasing sample rates. That is one of the reasons why the AudioRepair tool just works on 16 Bit / 44.1 kHz material. As a simple picture we could say that more samples just catch more of the analog wave recreated after the digital to analog conversion.
If the mastering engineer uses an oversampled limiter then this wouldn't be an issue at all, but unfortunately there are a lot of releases creating inter sample peaks by intention to be as loud as possible. If the DAC is able to handle the digital overs of up to +3 dB by its analog amplifier stage then the music would be really louder, but most of the available DACs just create distortions.

We wrote a White Paper about the reasoning behind High-Res-Audio. Maybe that provides a bit of an insight: https://www.xivero.com/blog/high-resolution-or-not-high-resolution-that-is-the-question/

Best regards,

Stephan


Offline Kempy

  • Newbie
  • *
  • Posts: 20
Dear Stephan,

thank you for confirming my results as features of the tool.

On the other hand I have started to realize how complex this topic is and how many factors are relevant to reproduce good sound. I need to need to deep dive here.

Best regards
Geozem

As Stephan told, AudioRepair is an other approach to handle ISP's in audio files, and imho the most ambitious.
On the downside, your have to store your files twice, the originals and the ISP corrected one.
Or, replace your originals with the ISP corrected.
I personally don't want that.

An further approach to handle ISP's in audio files is attenuating your playback level by -3.0dB in your playback app, preferably in the pre-amp section of the playback app.
You will cover most ISP's, and overshoots for lossy audio files too; your original files will remain untouched.
Downside, your playback will not be Bit-perfect. -3.0dB(=0.5bit) means SNR reduced by -3.0dB, not realy hard for 16bit material and pointless for 24bit sources.
To stay Bit-perfect you will need to attenuate by -6.0dB(=1bit) and if the playback app is well coded it will use bit-shifting for the -6.0dB attenuation.

Third way is to use an real-time oversampled True-Peak Brickwall Limiter with proper settings in the playback chain.
Hopefully it will alter/limit only levels above 0dB True-Peak, beside of it's oversampling-limiting-downsampling processing.

Always take care that the playback app is able to apply proper Dither to the final bit-depth.
Playback app's and DSP's are working in 32bit or 64bit Floating Point internaly.
For best sound, the Playback app should apply Dither before going to the final 16bit or 24bit Integer output bith-depth.
As far as I know, e.g. Foobar is able to dither to 16bit Int only in WASAPI mode.
If dithering to 24bit Int is realy required is a point to discuss, but I prefer proper dithering, to 24bit Int and 32bit Int too.

Regards

Offline geozem

  • Newbie
  • *
  • Posts: 36
Thank you for adding more background information to my toopic here.

I have decided to keep allways the original WAV/Flac files on a second NAS in case I need to redo any kind of correction later on.
But thank you for your advise here.

By playing around with Audicity I have found out that the fixed point scaling of the data is a must and is not that easy without the right tools and understanding of the  parameters. So as we are in the embedded world and not in the analog I can follow more your explanations now.
I am still learning.

On the other hand Dither is an interesting topic. If I see other tools they put it in without telling you as an option or they ignore it and then at the playback I have some DSP resampling inacuracies where it could stop the playback or jump to the next correct digit.

So the files converted by XiSRC are the stable ones. Runns without any issue.

Except the Nyquist Filter that you apply. I am still unhappy that this is be default not changable. I can understand the signal Theorem behind it as I have studied it as a mechanical engineer but my experience is that different DSPs has different design and therefore different result which at least I can clearly here.

Probably I need to learn more in this area to follow Stephans explanations to except the configuration as it is. Even to ignore my ears by this. The brain is telling different as the ear hears. Very difficult.

Best regards
geozem

Offline Kempy

  • Newbie
  • *
  • Posts: 20

On the other hand Dither is an interesting topic. If I see other tools they put it in without telling you as an option or they ignore it and then at the playback I have some DSP resampling inacuracies where it could stop the playback or jump to the next correct digit.

Best regards
geozem

A simple advice, if you are on Windows.
Prevent Direct Sound (DS) output mode in your playback app. Use ASIO or WASAPI or Kernel Streaming when possible.

Regards
« Last Edit: October 20, 2015, 01:38:18 PM by Kempy »

Offline geozem

  • Newbie
  • *
  • Posts: 36
Thank you for the further advise.

So far I am using Yamaha with included DLNA renderer to play my flac files from a NAS.

But I have used my Laptop with Jriver befor that so I can confirm your advise as best practice for best results.

best regards
geozem