Skip to main content
Topic: My result of Wavpack hybrid 483 Kbps vs lossless (Read 2704 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

My result of Wavpack hybrid 483 Kbps vs lossless

Spoiler (click to show/hide)

It's worth mentioning that I used the version 5.0.0 alpha 4.

Some might think that this is the 3rd ABX attempt with wavpack (by loocking at the files names) but nope, actually I did like 12 test with different files and I wasn't able to ABX any of those at 384 kbps (Test A, B, C, D, etc) So I started with the letters again and different files because I was seriously considering to convert all my Flac library to hybrid at said bitrate, unfortunately I ran into this one. xd All my previous +12 attempts were with "-b384x3" so I started to bump things one by one until I reached -hhb512x6 (Not very sure if the parameters are right), which gives a bitrate of 483 kbps.

Tip: Focus on 0:03 to 0:06, the hybrid's hiss sounds a bit different, it has a higher tone.

 

Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #1
Hi,

 You can try to disable resampling, use stable version, Try different noise shaping method: s0=off, s1=shift noise to hf
wavpack 4.8 -b256hx6c

Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #2
Ive tested it briefly using wv 4.60 which is same as 4.80.  At -b3x4 , -b4x4 using default noise shaping and s1. Volume is  higher than normal listening.

@ b3 (270k)  - obvious hiss around note but not bad.  Using S1 is worse than DNS shaping.
@ b4 (352k) - Much harder that I could not be bothered. I suspect maybe a tiny imperfection. Using S1 at this high bitrate didn't affect it and I had the impression it was even better, though I also didn't abx past 1/3
wavpack 4.8 -b256hx6c

Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #3
Ive tested it briefly using wv 4.60 which is same as 4.80.  At -b3x4 , -b4x4 using default noise shaping and s1. Volume is  higher than normal listening.

@ b3 (270k)  - obvious hiss around note but not bad.  Using S1 is worse than DNS shaping.
@ b4 (352k) - Much harder that I could not be bothered. I suspect maybe a tiny imperfection. Using S1 at this high bitrate didn't affect it and I had the impression it was even better, though I also didn't abx past 1/3


Version 4.80.0. without resampler (No noise shaping changes in this one).

Spoiler (click to show/hide)

And with -s1:

Spoiler (click to show/hide)


I also had the impression that S1 is better, I failed my first attempt with my superluxes 681 Evo, when this happens I usually give it a last shot with a pair of 668B (I call those the treble blasters xd) and managed to pass the test.




Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #4
I tried some more this evening. I managed -b300x4  5/5 and failed at x5 and 6. I thought -b320x3 was different but got 4/5.
As it is very tonal and that note is lower in volume , There is little masking and the noise could be heard - esp if artificially boosted in volume and EQ. In this case the default auto noise shaping gives it a 'rougher' more forward tone as i think it tries to shift the noise down in the frequency in order to hide it more but its still audible. I think this is what you mean by 'higher tone' and thats possibly attention grabbing as it may sound more like artifacts. Using no shaping or a static value gives it a more neutral character IMO. 

I am not too interested in boosting the volume to create a DSP effect as this defeats what would happen in normal listening. My impression is that at 300k or more (esp 384k) I don't think I would hear anything different.
wavpack 4.8 -b256hx6c

Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #5
I also did a side by side comparison in normal non-abx listening with -b256x4 and did not perceive anything different. Only when using -b200x4 it becomes obvious. I suppose with track replaygain it could become more obvious.
wavpack 4.8 -b256hx6c

Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #6
Thanks for reporting these results...very interesting! The first thing I did was run a spectrum analysis of the entire clip and immediately saw what the problem was. There is a +24 dB boost in the noise floor above 15 kHz! It's kind of pretty with that double helix, but I really can't imagine any good reason for it to be there, and it's causing a lot of trouble.



I created a new version with that boost removed, and by just comparing lossless compression rate it's costing 135 kbps, which also means it's also requiring about that amount additional in the hybrid mode. In fact, the "fixed" version compresses losslessly to 484 kbps!

Here's a plot of the spectrum of the left channel original audio (cyan) and the lossy version's added noise (magenta) from 0:03 to 0:06. There are a few places where the added noise gets less than 10 dB away from the noise floor, and that's probably what you're hearing. Using the "fixed" version, I was able to get that same noise performance at the 384 kbps setting (actual 360 kbps).



This is kind of a torture test for WavPack hybrid, and this is the kind of situation that would be solved with an “intelligent” VBR hybrid mode (it would simply detect this issue and boost the bitrate, or maybe even switch to lossless). Of course, it can be argued that no properly recorded or mastered recording would ever look like this, and certainly not from acoustic instruments, but that's obviously no excuse for WavPack.

Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #7
Thanks for reporting these results...very interesting! The first thing I did was run a spectrum analysis of the entire clip and immediately saw what the problem was. There is a +24 dB boost in the noise floor above 15 kHz! It's kind of pretty with that double helix, but I really can't imagine any good reason for it to be there, and it's causing a lot of trouble.



I created a new version with that boost removed, and by just comparing lossless compression rate it's costing 135 kbps, which also means it's also requiring about that amount additional in the hybrid mode. In fact, the "fixed" version compresses losslessly to 484 kbps!

Here's a plot of the spectrum of the left channel original audio (cyan) and the lossy version's added noise (magenta) from 0:03 to 0:06. There are a few places where the added noise gets less than 10 dB away from the noise floor, and that's probably what you're hearing. Using the "fixed" version, I was able to get that same noise performance at the 384 kbps setting (actual 360 kbps).



This is kind of a torture test for WavPack hybrid, and this is the kind of situation that would be solved with an “intelligent” VBR hybrid mode (it would simply detect this issue and boost the bitrate, or maybe even switch to lossless). Of course, it can be argued that no properly recorded or mastered recording would ever look like this, and certainly not from acoustic instruments, but that's obviously no excuse for WavPack.


Thanks for explaining what was going on.

Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #8
Hello David,

Would it be technically feasible to implant the vbr mode that was in 4.0alpha into 5 alpha ?  Would it affect anything adversely ?
wavpack 4.8 -b256hx6c

Re: My result of Wavpack hybrid 483 Kbps vs lossless

Reply #9
shadowking,

There are some differences in the way noise shaping was implemented then compared to now, so the decoder side would have to change, and I'm not willing to make that any more complicated than it is now. Any future vbr mode is going to have to be all new (and preferably not break old decoders).

 
SimplePortal 1.0.0 RC1 © 2008-2019