Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: A killer sample for Opus 256 VBR (Read 2804 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

A killer sample for Opus 256 VBR

Guys, I was playing around with opus (libopus 1.5.2) 256 kbps vbr (which sounded great to me so far) until I stumbled upon this one track where I immediately noticed a difference in the way drums sounded. I thought it was placebo at first but then I ran an abx in training mode (just for fun) and passed 5/5, twice :-\ . Then I ran an actual abx and failed but I swear I can hear the difference. Need people with better ears to listen to it.
Attached a 30 sec file

Spoiler (click to show/hide)

Re: A killer sample for Opus 256 VBR

Reply #1 –
Nothing, my man. To me, they sound the same. But maybe someone else could also test.
TAPE LOADING ERROR

Re: A killer sample for Opus 256 VBR

Reply #2 –
After 2 + 2 initial drum hits, when guitar and bass come in, listen carefully to drums  - they sound softer (less precise) in opus. The difference is subtle but it's there.
 

Re: A killer sample for Opus 256 VBR

Reply #3 –
I've got quite similar results on the first run. Listening window was from 2.5 sec  to 12.8 sec.

Total: 6/8
p-value: 0.1445 (14.45%)

Spoiler (click to show/hide)

I assume the cause for the difference are the clipped drumbeats in the original, the Opus version is missing them (see attached screenshot).

Re: A killer sample for Opus 256 VBR

Reply #4 –
Quote from: .halverhahn on
I've got quite similar results on the first run. Listening window was from 2.5 sec  to 12.8 sec.

Total: 6/8
p-value: 0.1445 (14.45%)

Spoiler (click to show/hide)

The thing is, drums on this track aren't even that pronounced, they're somewhat in the back. The accent in the mix is made on guitar riffs which are a lot louder and overwhelm my hearing quite fast so I can't reliably pass more trials.

Re: A killer sample for Opus 256 VBR

Reply #5 –
Quote from: .halverhahn on
I assume the cause for the difference are the clipped drumbeats in the original, the Opus version is missing them (see attached screenshot).



Interesting. The difference is too high in frequency though. I don't think this would be audible. The waveform has no clipping neither in the flac nor in the opus file.

Re: A killer sample for Opus 256 VBR

Reply #6 –
The sample is not full-scale clipped, but it has some quite similar square-wave-character samples in a row.
I'm just guessing this creates some click sound that accentuate the drumbeat.

Lets see what happens when listening to a 20kHz lowpassed version of the original.

Re: A killer sample for Opus 256 VBR

Reply #7 –
Quote from: .halverhahn on
Lets see what happens when listening to a 20kHz lowpassed version of the original.

Ok, sounds fine to me now)) This is weird, I thought opus did the lowpass filtering itself. Now we have to lowpass filter prior to encoding?😑

Re: A killer sample for Opus 256 VBR

Reply #8 –
Tried AAC 256 tvbr with this sample. Sounds transparent to me (albeit, having a little higher bitrate). I guess I'll be using that.
Shame though, opus doesn't screw up this sample because of its bad encoding technique, but because of its mandatory lowpass filtering at any bitrate. But no matter the reason, it doesn't deliver a transparent result.

Re: A killer sample for Opus 256 VBR

Reply #9 –
Quote from: Teddy_the_barber on
Ok, sounds fine to me now)) This is weird, I thought opus did the lowpass filtering itself. Now we have to lowpass filter prior to encoding?😑

My intention was to compare the original vs the lowpassed FLAC version. May you hear a difference - I currently have a listening fatigue on this one - please give me some time.

Re: A killer sample for Opus 256 VBR

Reply #10 –
Quote from: .halverhahn on
Quote from: Teddy_the_barber on
Ok, sounds fine to me now)) This is weird, I thought opus did the lowpass filtering itself. Now we have to lowpass filter prior to encoding?😑

My intention was to compare the original vs the lowpassed FLAC version. May you hear a difference - I currently have a listening fatigue on this one - please give me some time.


Yepp, still hear it. The sound of drums in the original is more "piercing". I feel like I got honest 6/8 though, the last one was hard and likely could be a guess.

Spoiler (click to show/hide)

Re: A killer sample for Opus 256 VBR

Reply #11 –
It was quite hard for me, but I've managed to differentiate originals from the lowpassed FLAC version.

So, the difference is not the Opus encoding, its the lowpass. As it seems the lowpass smooths out square-wave-character samples, so the click sound that accentuate the drumbeat is gone.

Spoiler (click to show/hide)

Re: A killer sample for Opus 256 VBR

Reply #12 –
I guess you're right. Thanks for helping to clear this up.

Re: A killer sample for Opus 256 VBR

Reply #13 –
what kind of lowpass filter is this? is it sharp enough / if you subtract the files one from another, do they cancel out to silence below 20kHz?
if it is and if there is no clipping involved at any stage, it is very surprising that one can hear a difference above 20kHz.
(to really make sure there's no clipping, you could just make them both quieter by e.g. some -6 dB, and just turn the playback volume as necessary when testing)
could it also be some distortion of those high frequencies leaking into a (more) audible range? what happens if you just play back the difference that only has >20kHz part?
a fan of AutoEq + Meier Crossfeed

Re: A killer sample for Opus 256 VBR

Reply #14 –
Quote from: magicgoose on
[...] it is very surprising that one can hear a difference above 20kHz.

Quote from: magicgoose on
what happens if you just play back the difference that only has >20kHz part?
Not doubting the great assessment done so far by @.halverhahn , but those were just my thoughts: how come??


• Listen to the music, not the media it's on
• The older, the 'lossier'

Re: A killer sample for Opus 256 VBR

Reply #15 –
Quote from: magicgoose on
what kind of lowpass filter is this? is it sharp enough / if you subtract the files one from another, do they cancel out to silence below 20kHz?
I've used the Sox Lowpass filter created by Case for foo2k https://foobar.hyv.fi/?view=foo_dsp_lowpass.
Used settings are: Lowpass: 20000hz, lowpass width: 5%, Phase response: linear
Feel free to create your own lowpassed version using the sample provided by @Teddy_the_barber and compare/nulling it to my lowpass20k.flac sample.

Btw. Yes, they subtract one from another blow 20kHz.

Quote from: magicgoose on
if it is and if there is no clipping involved at any stage, it is very surprising that one can hear a difference above 20kHz.
It's also stange to me, as my hearing stops approximately at 16.5kHz (at insane listening levels).

Re: A killer sample for Opus 256 VBR

Reply #16 –
Quote from: magicgoose on
could it also be some distortion of those high frequencies leaking into a (more) audible range?
We're definitely not listening to a >20 kHz range here))  Opus sounds transparent if encoded from the lowpassed flac, while opus file and the lowpassed flac can be abx`ed against the original flac. So the only explanation is as @.halverhahn rightly assumed- some sort of distortion from frequencies above 20 kHz bleeds into the (very much) audible range.

Re: A killer sample for Opus 256 VBR

Reply #17 –
Quote from: Teddy_the_barber on
Quote from: magicgoose on
could it also be some distortion of those high frequencies leaking into a (more) audible range?
We're definitely not listening to a >20 kHz range here))  Opus sounds transparent if encoded from the lowpassed flac, while opus file and the lowpassed flac can be abx`ed against the original flac. So the only explanation is as @.halverhahn rightly assumed- some sort of distortion from frequencies above 20 kHz bleeds into the (very much) audible range.
You mean in a harmonics/intermodulation sort of way? 
Doesn't that sound like aliasing's playing a role in it?
• Listen to the music, not the media it's on
• The older, the 'lossier'

Re: A killer sample for Opus 256 VBR

Reply #18 –
Quote from: includemeout on
You mean in a harmonics/intermodulation sort of way?
It could very well be it. Looks to me like this content above 20 kHz introduces intermodulation distortion into the audible range at the stage of actual D/A conversion in a dac (the file itself doesn't contain anything like that as everything below 20 kHz is substracted perfectly from the lowpassed track). Filtering the frequencies above 20 kHz removes the problem as there's nothing to trip up the dac anymore. Imo, that track probably sounds more "proper" after it's filtered.
Anyway, an interesting sample, I think.

Re: A killer sample for Opus 256 VBR

Reply #19 –
Yes that's what I would think, too. Aliasing or nonlinear distortion somewhere in the playback chain after the decoding.
A similar, more extreme case can happen when playing audio sourced from a SACD (DSD) which hasn't been lowpassed, there is so much ultrasound noise.

> Sox 20000hz, lowpass width: 5%, Phase response: linear

Yeah that should be good enough by far.
a fan of AutoEq + Meier Crossfeed

Re: A killer sample for Opus 256 VBR

Reply #20 –
I have been meaning to ask if you who hear difference could share what you use to listen to the files. The lowpassed file shared here looks perfect, the difference between it and the original only contains very quiet high frequency tones. It's completely inaudible to me even using headphones at loud volume. The signals are nowhere loud enough to cause clipping, even intersample peaks are about 1.3 dB below digital fullscale.
My logical mind says the files should sound the same, unless the difference signal is also somehow audible for you.

X

Re: A killer sample for Opus 256 VBR

Reply #21 –
Quote from: Case on
I have been meaning to ask if you who hear difference could share what you use to listen to the files.
You mean equipment I'm using?

Quote from: Case on
The lowpassed file shared here looks perfect, the difference between it and the original only contains very quiet high frequency tones. It's completely inaudible to me even using headphones at loud volume.

Same here. Played the diff track - silence.

Quote from: Case on
My logical mind says the files should sound the same
Logically they should. They don't though. As I've already mentioned previously "Opus sounds transparent if encoded from the lowpassed flac, while opus file and the lowpassed flac can be abx`ed against the original flac."

Re: A killer sample for Opus 256 VBR

Reply #22 –
Quote from: Teddy_the_barber on
Quote from: Case on
I have been meaning to ask if you who hear difference could share what you use to listen to the files.
You mean equipment I'm using?
Yes, though I was especially curious about equipment that causes audible difference between the two FLACs. Hoping to find measurements for the gear that shows that there indeed is distortion.

Re: A killer sample for Opus 256 VBR

Reply #23 –
Mine is Modi+ connected to a pair of Adam T5V's

Re: A killer sample for Opus 256 VBR

Reply #24 –
Quote from: Teddy_the_barber on
Quote from: Case on
The lowpassed file shared here looks perfect, the difference between it and the original only contains very quiet high frequency tones. It's completely inaudible to me even using headphones at loud volume.
Same here. Played the diff track - silence.

+1 - Pure silence for the diff track.

My gear for this ABX is a Dell Latitude 7240 headphone output (Realtek Audio), Windows 11, Windowsmixer set to 24/48000, using cheap Panasonic RP-TCM115E In-Ear Headphones.