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: Opus 1.3-rc (Read 53695 times) previous topic - next topic
0 Members and 2 Guests are viewing this topic.

Re: Opus 1.3-rc

Reply #25
Now this is interesting. Could you describe the kind of distortion which is happening here? Maybe it's an encoder bug that can be fixed.
And did you use ReplayGain to avoid clipping while testing? The Opus file has peak level ≈1.166; even though clipping is less likely to be noticeable behind this wall of noise, there's a chance someone can notice it.
a fan of AutoEq + Meier Crossfeed

Re: Opus 1.3-rc

Reply #26
I would imagine we are talking very small differences in sound quality to where when casually listening I would not expect to hear any obvious differences between v1.2.1 and v1.3RC at bit rates of 32-48kbps (give or take).
Here we don't imagine things but actually  test it.

1.3RC is significantly better than 1.2.1 at least for speech  at low bitrates. Not only 1.3RC is much better than 1.2.1 but also bitrates are less bloated on speech samples.

P.S. 1.3RC has new advanced speech detector based on algorithms of artificial intelligence, RNN.

Re: Opus 1.3-rc

Reply #27
--vbr --bitrate 180

foo_abx 2.0.4 report
foobar2000 v1.4 beta 17
2018-06-08 13:41:39

File A: precision010.flac
SHA1: a1b28d8473317a6ee814598e9542d3aa63e0472f
File B: precision010.opus
SHA1: 8ae8807b027516fb9748b710f7451c4cce005535

Output:
DS : Первичный звуковой драйвер
Crossfading: NO

13:41:39 : Test started.
13:42:49 : 01/01
13:43:30 : 02/02
13:43:50 : 03/03
13:44:24 : 04/04
13:45:05 : 05/05
13:45:21 : 06/06
13:45:39 : 07/07
13:45:57 : 08/08
13:46:22 : 09/09
13:46:32 : 10/10
13:46:55 : 11/11
13:47:15 : 12/12
13:47:34 : 13/13
13:47:59 : 14/14
13:48:18 : 15/15
13:48:53 : 16/16
13:48:53 : Test finished.

 ----------
Total: 16/16
Probability that you were guessing: 0.0%

 -- signature --
bdb235586eed63116588b7b4dd615c7ddfb6a855


my audio hardware: DAC ES9018K2M + TPA6133A2

Interesting. What headphone do you have for the test?

Most of my library is converted using Opus 1.2.1 at 176kbps, I should hope that 1.3 didn't improve too much on the higher birates :)

Re: Opus 1.3-rc

Reply #28
I would imagine we are talking very small differences in sound quality to where when casually listening I would not expect to hear any obvious differences between v1.2.1 and v1.3RC at bit rates of 32-48kbps (give or take).
Here we don't imagine things but actually  test it.

1.3RC is significantly better than 1.2.1 at least for speech  at low bitrates. Not only 1.3RC is much better than 1.2.1 but also bitrates are less bloated on speech samples.

P.S. 1.3RC has new advanced speech detector based on algorithms of artificial intelligence, RNN.

I did notice a tiny difference in file size between v1.3RC and v1.2.1 on the sample (i.e. 24-Greensleeves-Korean-male-speech.441) you linked to...

v1.2.1 (@ 16kbps) =19.5 KB (19,974 bytes)
v1.3RC (@ 16kbps) = 19.2 KB (19,682 bytes)

but other than that, I can't say I notice any obvious difference between the two when listening to those two Opus speech files, at least based on my Klipsch Pro-Media speakers for the PC. if there is a difference it's got to be minimal especially given that low of a bit rate as I set it to 16kbps and v1.3RC shows 17kbps and v1.2.1 shows 18kbps in Foobar2000.

so your saying Opus v1.3's greatest advantage over v1.2.1 is definitely speech? ; if that's true, I would not expect to hear any obvious difference between v1.3 and v1.2.1 if that speech file you linked to is a good test for the encoders. but who knows, maybe there could be differences in other random speech files(?).

but I just noticed that there is some volume change between the WAV file and the Opus files and I know what the WAV sounds like so it's easy for me to detect the WAV vs Opus files on that alone when ABXing. like in the WAV file, the left speaker is louder where as with the Opus v1.2.1 and v1.3RC those sound the same in that regard in that the left speaker has less volume and is quieter. like I can adjust volume (on the speaker itself) so it's nice and even on both speakers with the Opus files but when I play back the WAV file the left speaker is noticeably louder. but just strictly playing the v1.2.1 and v1.3RC and listening, back and fourth, I can't say I feel confident enough to say one sounds noticeably better than the other as they seem pretty much the same to me. but trying to overlook the volume change, I am pretty sure I could ABX those 16kbps Opus speech files from the WAV. but in terms of just doing a straight compare of v1.2.1 and v1.3RC, I don't feel confident enough to say I hear any clear difference between the two files.
For music I suggest (using Foobar2000)... MP3 (LAME) @ V5 (130kbps). NOTE: using on AGPTEK-U3 as of Mar 18th 2021. I use 'fatsort' (on Linux) so MP3's are listed in proper order on AGPTEK-U3.

Re: Opus 1.3-rc

Reply #29
but other than that, I can't say I notice any obvious difference between the two when listening to those two Opus speech files, at least based on my Klipsch Pro-Media speakers for the PC.
ah, You listen through speakers which are less sensitive to artifacts. That explains the most part of your results.

Headphones are much more revealing when it comes to ABX&ABCHR tests

For example,
http://listening-tests.hydrogenaud.io/sebastian/mp3-128-1/index.htm
Quote
... Headphones are a must-have

Re: Opus 1.3-rc

Reply #30
ah, You listen through speakers which are less sensitive to artifacts. That explains the most part of your results.

Headphones are much more revealing when it comes to ABX&ABCHR tests

I see. but I think I can safely say my speakers are above average computer speakers, if that matters at all. because I could easily see how if I had lower quality speakers (say pretty basic/average range) it would be more difficult to spot artifacts etc. basically all comments I have made around here are generally based around those Klipsch Pro-Media speakers which I have had since about the early 2000's. I got a couple of old ABX logs from the year 2013 testing MP3 @ v5(130kbps) and v6(115kbps) on a random song and I pretty much successfully ABX'ed em on my speakers, if that matters at all. basically I got... 11/12 (0.3%) for the v5 (130kbps) and 10/10 (0.1%) for the v6 (115kbps) if that helps give you any rough indication of their ability for artifact spotting etc.

but with that said... when it comes to headphones, do I need anything fancy? ; because all I have is Sony MDR-NC7 headphones which are probably not crap(as I would assume it could be worse), but definitely nothing special.

either way, with speakers or headphones... that would not explain the left and right speaker volume issue I had when testing that sample you linked to in here, correct? ; that's the only thing I noticed that issue on as with other random speech files I got and general music it does not have that volume issue from what I noticed.

but would it be possible for me to detect artifacts on those speakers but not be able to on the headphones? ; although I realize as a general guideline, based on what your saying, headphones are preferred in general.

just some thoughts ; but I guess in the meantime we will just have to assume your tests are good with v1.2.1 vs 1.3RC ;)
For music I suggest (using Foobar2000)... MP3 (LAME) @ V5 (130kbps). NOTE: using on AGPTEK-U3 as of Mar 18th 2021. I use 'fatsort' (on Linux) so MP3's are listed in proper order on AGPTEK-U3.

Re: Opus 1.3-rc

Reply #31
Interesting. What headphone do you have for the test?
Uiisii hm7 + rubber ear Pads
Could you describe the kind of distortion which is happening here?
Good to hear the hiss in intense fields, (if apply by sinc interpolation (foo_dsp_multiresampler) resampling of 44.1k - 48k hiss becomes audible). Heard aggressive removal of side (L-R / 2). At the moment I am very pleased with the work of only FAAC 1.28 (-q160) it is great.
And did you use ReplayGain to avoid clipping while testing?
Nothing was used

Re: Opus 1.3-rc

Reply #32
Could you describe the kind of distortion which is happening here?
Good to hear the hiss in intense fields, (if apply by sinc interpolation (foo_dsp_multiresampler) resampling of 44.1k - 48k hiss becomes audible). Heard aggressive removal of side (L-R / 2). At the moment I am very pleased with the work of only FAAC 1.28 (-q160) it is great.
And did you use ReplayGain to avoid clipping while testing?
Nothing was used
If you add some effects (a resampler which audibly changes sound?!) then the ABX test is not correct.
re. Aggressive removal of side (L-R / 2): if you listen to the channel difference in isolation (by shorting 2 of 3 wires in  headphones, or otherwise) this is also not a fair test. (Or can you isolate (L-R) by ear so much better than other people?)
Lossy encoders are normally optimized for listening without any drastic effects like these.

ReplayGain, OTOH, should be used for comparison to prevent clipping and/or loudness difference. (Unless you're testing a codec which has an additional goal of not increasing peak values; Opus has no such design goal, neither does AAC or other typical lossy codecs)
a fan of AutoEq + Meier Crossfeed

Re: Opus 1.3-rc

Reply #33
magicgoose, This ABX test was done by ear, and I already wrote that I'm not a professional and can be wrong. The reasons that I described are subjective but indicate a problem at high bitrates.

Re: Opus 1.3-rc

Reply #34
If it was done by ear entirely, then why do you mention some postprocessing which, allegedly, alters sound in interesting ways?
I don't really understand why it's relevant in this case.
Quote
if apply by sinc interpolation (foo_dsp_multiresampler) resampling of 44.1k - 48k hiss becomes audible
Was this resampler used during the test?
a fan of AutoEq + Meier Crossfeed

Re: Opus 1.3-rc

Reply #35
but with that said... when it comes to headphones, do I need anything fancy?
If you're comfortable with your hardware and sound then  there is no need to bother.

I see. but I think I can safely say my speakers are above average computer speakers...
It's not about speaker's price (well, at least it's not only about it) but a general trend that  good pair of headphone reveals artifacts of lossy compression better.

Re: Opus 1.3-rc

Reply #36
Resampling is unavoidable due to the fact that opus is fixed at 48kHz and the original file is 44.1kHz. I looked into the log and it was done by using DS. Skipping the resampler plugin and Windows built-in resampler will still resample.

kode54 mentioned that foo_dsp_multiresampler's quality is not state of the art, but with the sinc option it is still transparent to me.
https://hydrogenaud.io/index.php/topic,112883.msg929567.html#msg929567

The potential clipping/DS limiting issue on the other hand should be addressed.

 

Re: Opus 1.3-rc

Reply #37
If it was done by ear entirely, then why do you mention some postprocessing which, allegedly, alters sound in interesting ways?
I don't really understand why it's relevant in this case.

I didn't use anything to prove inside the encoder tags the original sample rate was listed. no interpolation was used in the test and replaygain was also not used. you're using my words against me...

Was this resampler used during the test?
I didn't use anything!
you can encode the attached flac file in opus and make a bitwise comparison with what I provided.

Quote
if apply by sinc interpolation (foo_dsp_multiresampler) resampling of 44.1k - 48k hiss becomes audible

(if apply) does not mean applied.
I wrote it thinking it could theoretically help.

The potential clipping/DS limiting issue on the other hand should be addressed.
I agree with you


Re: Opus 1.3-rc

Reply #39
https://hydrogenaud.io/index.php/topic,114871.msg946989.html#msg946989

According to Arnold's comment ES9018K2M is potentially problematic with filtering, sadly he didn't provide the recorded files. The aliasing of HF content could potentially affect ABX results as well.
You not have evidence to support the presence of distortion in ES9018K2M, you can ask a question in the support service: http://www.esstech.com/index.php/en/support/support-contact/

Re: Opus 1.3-rc

Reply #40
You not have evidence to support the presence of distortion in ES9018K2M
Yeah, and while you are reading the post, be careful with the attached test signals. When used improperly they can harm your equipment and hearing.
Quote
you can ask a question in the support service
In case you don't know, those ESS chips have configurable digital filters, a DAC chip is different from the whole product with all other circuitry and the manufacturer can also reprogram the chip to change its behavior. Without the actual audio file to examine with asking ESS themselves is useless. ESS even don't allow everyone to download their datasheet publicly, you can definitely find some with the help of Google and the pdf files are watermarked with "confidential" on every page. They will not disclose everything to the public openly.

Re: Opus 1.3-rc

Reply #41
I tested it from windows and there were no apparent bugs.

Encoding on 9kbps-13kbps range was a bit slower than the other rates.
I don't believe this is a serious issue, but anyway.

Other than the bitrate, the default settings were used.
Code: [Select]
opus-tools-test-1.3-rc\opusenc --bitrate %v %i %o
What does %v %i %o mean/do? I'm really curious about that.

Re: Opus 1.3-rc

Reply #42
...What does %v %i %o mean/do? I'm really curious about that.

Those are variables to his program or batch file. In this case meaning BitrateValue InFile OutFile respectively.

Re: Opus 1.3-rc

Reply #43
Here are few outdated results from 1.3 post-beta testing.

https://drive.google.com/file/d/1eusrdEWxtg8ya8RkMl8v_17FbHMfkWBv/view
50% of samples speech, 50% music

MOS (32 kbps VBR, stereo):
version 1.2.1 - 3.35
version 1.3 post-beta1  B (2c0d9ee) - 3.98

The most important part of improvements of 1.3 goes for speech and mixed material.


Korean speech 16-40 kbps
https://drive.google.com/file/d/0ByvUr-pp6BuUOHBIU2dXaVNZU0tMblRyU3hnU3llSHRaVDMw/view?usp=sharing

A opus-tools-cdcb2f7
B opus-tools-2c0d9ee
C opus-tools-b44f56e
D opus-tools-9cfdcba

These A, B, C, D are builds  in-between 1.3b1 and 1.3RC1

German speech sample https://drive.google.com/file/d/1SnAd_vEpfT_ndF0cmqlQ2LpwchNMNTbS/view?usp=sharing
1.2.1 - 3.0
1.3RC1 - 3.5

P.S. I would love to test something more complete and solid but it's not right time for me. Hopefully later.

Re: Opus 1.3-rc

Reply #44
Nice improvements!

Re: Opus 1.3-rc

Reply #45
Sorry if this is a noob question, but would this update be worth replacing my current library of opus 1.3 beta encoded at 160kbps? Thanks in advance.

Re: Opus 1.3-rc

Reply #46
Sorry if this is a noob question, but would this update be worth replacing my current library of opus 1.3 beta encoded at 160kbps? Thanks in advance.

I'm sure someone will correct me if i'm totally wrong here, but, to answer your question: Almost certainly not. I'd just start using 1.3 when it is released for your new encodes from that point onward... at those bitrates the improvements (if any) will be very hard to notice.

Re: Opus 1.3-rc

Reply #47
if apply by sinc interpolation (foo_dsp_multiresampler) resampling of 44.1k - 48k hiss becomes audible

I do not understand this part. The opus file is already 48khz. Resampling to 48khz shouldn't do anything.

Are you sure you are not artificially introducing artifacts in your listening setup somehow? I love Perturbator and tried to ABX your files but couldn't spot a difference (it's quite difficult anyway because it's already so heavily distorted).

Re: Opus 1.3-rc

Reply #48
Sorry if this is a noob question, but would this update be worth replacing my current library of opus 1.3 beta encoded at 160kbps? Thanks in advance.

I'm sure someone will correct me if i'm totally wrong here, but, to answer your question: Almost certainly not. I'd just start using 1.3 when it is released for your new encodes from that point onward... at those bitrates the improvements (if any) will be very hard to notice.
Much appreciated, thank you.

Re: Opus 1.3-rc

Reply #49
Sorry if this is a noob question, but would this update be worth replacing my current library of opus 1.3 beta encoded at 160kbps? Thanks in advance.

Like Deathcrow already said, it would be almost certainly a waste of time simply because @ 160kbps the wiki page says, 160-192kbps = "Transparent with very low chance of artifacts (a few killer samples still detectable)." ; so when your already borderline perfection, it's not really possible for v1.3 to improve on that to any real degree even though I guess technically if it fixed those rare killer samples, that would be a improvement but at this point you would be nitpicking for sure.

basically with v1.3 it seems the sound quality improvements are mainly at bit rates of around 48kbps and lower from what I have read around here. but on the higher bit rates (say about 96kbps+) I would expect to see very little if anything since they are already high quality on the current release of v1.2.1 as 96kbps shows 'approaching transparency' on the wiki page for Opus which means even that bit rate already offers quality sound. you can't really go wrong with 96kbps/128kbps/160kbps.

or let me bottom line Opus... I think I can say that the vast majority of people who have a concern for sound quality will use a bit rate of between 96-160kbps (maybe 192kbps for the somewhat paranoid types) as while you can go lower than 96kbps you start to gamble a bit on sound quality even though at the lower rates the sound quality is quite respectable given the bit rates, like... 32kbps/48kbps/64kbps etc. but any higher than 160-192kbps is pretty much a waste of storage space given the info in the Opus wiki page.
For music I suggest (using Foobar2000)... MP3 (LAME) @ V5 (130kbps). NOTE: using on AGPTEK-U3 as of Mar 18th 2021. I use 'fatsort' (on Linux) so MP3's are listed in proper order on AGPTEK-U3.