Hello guys.

I´ve did a search for this but I found no answer. I´m using WavPack a lot recently on high definition files (24/96) and I have a question. I would like to know why WavPack disables the dynamic noise shaping when used with sampling rates above 64 kHz?

Will I get a benefit when I use the switch --use-dns? When using WaveLab´s frequency analysis on files that were processed without any switch I´m getting dynamic high frequency content over 20 kHz which I assume is quantization noise. When I´m forcing the -dns switch this noise is much lower but I think it is still there. Where is it? Is it masked in the lower frequency regions (20-20.000 Hz)?

Thanks in advance for an answer. BTW, great codec!

Does this even make sense? WavPack is lossless, so the only effect of different switches would to be to change the packing efficiency, not to change the audio content.

I´m sorry, I forgot to mention that I always use the lossy (hybrid) compression due to limited harddrive space.

Ah, now I understand.

If you are using it lossy then is there any advantage to leaving it as 24/96?

When having original 24/96 files, yes. I don´t like to resample and other codecs like AAC or WMA-Professional simply discard everything above 20 kHz. I would like to keep them & save space at the same time. The lossless compression eats away too much of my limited space.

So do your ears

If you're not doing anything with those 24/96 files other than simple playback and you want smaller file-size, have you considered resampling and dithering down to 44.1/16? The bonus of doing that is that you can easily burn them to CD as well. You're not going to hear the difference if all you're doing is playing them back.

Please please please, no discussion over the benefit of 24/96 vs. 16/44.1. I just would like to know what happens to the quantization noise with high sampling rates when I use the dynamic noise shaping feature - nothing else.



Yes, I considered it but I listen only over my computer. I don´t burn files to CD (my CD player is barely used) and I don´t want to use a smaller sampling rate if the 24/96 files are original, say something from Linn or HDTracks.com. Please don´t let us discuss if one can hear 24/96 or not, this has been discussed on this forum before - with huge proportions.

I will post it again:
- I still would like to know where the noise goes when --use-dns is used

Please?

In a very simplified description, the dynamic noise shaping feature attempts to move the quantization noise toward where the signal is. At normal sampling rates (where all frequencies represented are potentially audible) this can reduce the audibility of the quantization noise because it is masked by the signal.

However, in the case of higher sampling rates where the top octave (and sometimes more) are beyond the range of human hearing, it might make more sense to bias the quantization noise in that direction, which can dramatically reduce the noise in the audible range. That's the reason that the dynamic noise shaping is disabled for high sampling rates; instead the noise shaping is equivalent to -s1.0 to force the noise up in frequency at 6 dB / octave. If you force the dynamic shaping on with the –use-dns switch then the noise will be permanently shifted toward lower frequencies because that's where the signal is.

I discuss this some in this thread and included some spectrum analysis showing the noise distribution. I did not do the same thing with DNS, but the noise in the audible range would be much higher.

BTW, thanks!

edit: added sentence for clarity

That´s it! Man, I love you. Thank you so very much, this answers my question exactly. Also spectrum analysis made everything clear immediately. Simple and perfect explanations.
As far as I can tell (from my amateurs´ point of view) you did a great job of creating this codec & using noise shaping. To my knowledge yours is the only codec that does noise shaping which I think is a very intelligent and unique approach for a lossy codec. I´ve used WMA-Professional before but now I can completely move over to WavPack.

Thanx again.