Skip to main content
Topic: Opus 1.3-beta is here (Read 27580 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Opus 1.3-beta is here

This beta release of the upcoming Opus 1.3 includes:
  • Enabling by default the spec fixes in RFC 8251
  • Improvements to the VAD and speech/music classification using an RNN
  • Improvements to stereo speech coding at low bitrate
  • Added support for ambisonics projection using mapping 3 (disabled by default)
  • Fixes to the CELT PLC
Additionally, as a way to test the upcoming update to opus-tools, we’re providing Windows binaries built with 1.3-beta. These binaries are based on libopusenc, which means opusenc is finally able to make use of the Opus delayed-decision feature to make better speech/music transitions.

Source code: opus-1.3-beta.tar.gz
Win32 binaries (experimental): opus-tools-test1.zip

Re: Opus 1.3-beta is here

Reply #1
Nice!

Any changes in fidelity in the 80-128 kbit range?

Re: Opus 1.3-beta is here

Reply #2
Thank you very much for such great a format!

From someone for whom Opus has become the exclusive low-bitrate choice for his (almost daily) walks, given that his ancient Moto X phone doesn't have an SD-card slot and storage space is therefore, always at a premium.
Listen to the music, not the media.


Re: Opus 1.3-beta is here

Reply #4
Nice!

Any changes in fidelity in the 80-128 kbit range?
the quality is already transparent for 2 channel stereo music at 48 kbps, but I wonder if they can make 5.1 or 7.1 audio sound transparent at 96 kbps, because in my tests, its transparent for 5.1 sound at 128kbps. I can't wait to see what's in store for the next year when it comes to efficiency at even lower bitrates than 48kbps!



Re: Opus 1.3-beta is here

Reply #7
the quality is already transparent for 2 channel stereo music at 48 kbps
Now is a good time for TOS 8,

Actually, TOS 8 is more about the opposite case -- people claiming they hear a difference between two signals. Anyone can produce the null hypothesis in an ABX test (through bad hearing, not caring, or just answering randomly) even when there's an obvious difference. OTOH, not everyone can produce significant results when two files are nearly identical. IOW, there's no point in saying "prove it" to someone claiming something is transparent to them.

Re: Opus 1.3-beta is here

Reply #8
There are some improvements ( "Can't wait" and "Fatboy" samples) but also some regressions. Most notable regression is on sample 4º.

I personally cannot hear much difference for sample #4, so I'm curious what artefact is annoying you. Is it something at a specific time?

Re: Opus 1.3-beta is here

Reply #9
It's the same artifact for all 3 samples (Korean speech, French speech, and Sample #4) where 1.3 does worse than 1.2.1. It's extra sibilance which possibly maybe described as "ringing", "paper sound" etc...

It repeats in a form of constant large periods during all time. It's not at a specific time.

This artifact is  present in L channel on the sample of Korean speech. Maybe it's stereo savings which cause this issue

Edit: grammatical errors

Re: Opus 1.3-beta is here

Reply #10
Just did some random transcode test with a song to opus format at low bitrate around 20kbps, i think Opus 1.3-beta perform worse than 1.2.1 at these low bitrate. i can hear some noise/artifacts that do not present in version 1.2.1-28. Suspect the speech/music detector mess up at these low bit rate.

Re: Opus 1.3-beta is here

Reply #11
Just did some random transcode test with a song to opus format at low bitrate around 20kbps, i think Opus 1.3-beta perform worse than 1.2.1 at these low bitrate. i can hear some noise/artifacts that do not present in version 1.2.1-28. Suspect the speech/music detector mess up at these low bit rate.

20kbps is way too low for music, maybe in a year or two it will reach the same quality as 48 kbps and up.

Re: Opus 1.3-beta is here

Reply #12
Thank you very much for such great a format!

From someone for whom Opus has become the exclusive low-bitrate choice for his (almost daily) walks, given that his ancient Moto X phone doesn't have an SD-card slot and storage space is therefore, always at a premium.


what is the bitrate that you encode your music in?

Re: Opus 1.3-beta is here

Reply #13
Here is my ABX of opus 1.3 using 48kbps and a rock track. Very easy to hear the difference and the opus 48kbps sounds "less focused" in the highs and sounds mono every now and then. I had to use an online ABX because my Foobar2000 under Arch Linux and WINE does not want to output any sound.

Re: Opus 1.3-beta is here

Reply #14
When I encoded the track at one time the opusenc produced weird result in the beginning. I had to convert it to FLAC from opus since the file I encoded was longer than thirty seconds and when I cut everything from 00:30 and onwards I thought that I might as well make it a FLAC.

Re: Opus 1.3-beta is here

Reply #15
When I encoded the track at one time the opusenc produced weird result in the beginning. I had to convert it to FLAC from opus since the file I encoded was longer than thirty seconds and when I cut everything from 00:30 and onwards I thought that I might as well make it a FLAC.

make sure that you are using libopus 1.2.1 dirty. that artifact in the beginning is a relic from opus 1.2 beta and the final version of 1.2. i too noticed that artifact, which was not present in the previous opusenc . i emailed jmvalin about that issue and he fixed it and released libopus 1.2.1

if you use opus 1.2.1 dirty, it will no longer have that artifact

Re: Opus 1.3-beta is here

Reply #16
When I encoded the track at one time the opusenc produced weird result in the beginning. I had to convert it to FLAC from opus since the file I encoded was longer than thirty seconds and when I cut everything from 00:30 and onwards I thought that I might as well make it a FLAC.

do you have dbpoweramp? well you can read tags  and see exactly what version of libopus it was encoded in

Re: Opus 1.3-beta is here

Reply #17
Here is my ABX of opus 1.3 using 48kbps and a rock track. Very easy to hear the difference and the opus 48kbps sounds "less focused" in the highs and sounds mono every now and then. I had to use an online ABX because my Foobar2000 under Arch Linux and WINE does not want to output any sound.


download libopus 1.2.1 dirty and transcode that flac into opus, and you will not hear any artifacts at 48 kbps. i have thousands of songs in this format at that bitrate and i never had problems with that revision.

Re: Opus 1.3-beta is here

Reply #18
here are some songs in 48 kbps  libopus 1.2.1

Re: Opus 1.3-beta is here

Reply #19
Here is my ABX of opus 1.3 using 48kbps and a rock track. Very easy to hear the difference and the opus 48kbps sounds "less focused" in the highs and sounds mono every now and then. I had to use an online ABX because my Foobar2000 under Arch Linux and WINE does not want to output any sound.


here is that same song you posted, but encoded in 1.2.1 dirty

it sounds pretty transparent to me

Re: Opus 1.3-beta is here

Reply #20
When I encoded the track at one time the opusenc produced weird result in the beginning. I had to convert it to FLAC from opus since the file I encoded was longer than thirty seconds and when I cut everything from 00:30 and onwards I thought that I might as well make it a FLAC.

make sure that you are using libopus 1.2.1 dirty. that artifact in the beginning is a relic from opus 1.2 beta and the final version of 1.2. i too noticed that artifact, which was not present in the previous opusenc . i emailed jmvalin about that issue and he fixed it and released libopus 1.2.1

if you use opus 1.2.1 dirty, it will no longer have that artifact
This artifact happened with b1.3 when I encoded the song but then didn't happen again after that. Really strange that it happened once but not again.

Re: Opus 1.3-beta is here

Reply #21
ehh.... what is opus 1.2.1 dirty

Re: Opus 1.3-beta is here

Reply #22
Here is my ABX of opus 1.3 using 48kbps and a rock track. Very easy to hear the difference and the opus 48kbps sounds "less focused" in the highs and sounds mono every now and then. I had to use an online ABX because my Foobar2000 under Arch Linux and WINE does not want to output any sound.


here is that same song you posted, but encoded in 1.2.1 dirty

it sounds pretty transparent to me

Interesting that you found that song on another album because I thought it was quite rare. Anyway, the track is full of artifacts which makes me wonder what age you are and if you have suffered any ear damage. I promise, I am not at all being rude but I felt I had to ask since the quality is bad and you say it is transparent to you. Regards.

Re: Opus 1.3-beta is here

Reply #23
the quality is already transparent for 2 channel stereo music at 48 kbps
Now is a good time for TOS 8,

Actually, TOS 8 is more about the opposite case -- people claiming they hear a difference between two signals. Anyone can produce the null hypothesis in an ABX test (through bad hearing, not caring, or just answering randomly) even when there's an obvious difference. OTOH, not everyone can produce significant results when two files are nearly identical. IOW, there's no point in saying "prove it" to someone claiming something is transparent to them.
The rule says to rule out subjective opinions and must therefore mean any opinion, transparent or not transparent. If that makes it difficult to know when someone just don't care which result in an ABX that says that the person could not distinguish between to files is then up the Hydrogenaudio to sort out. I can talk you into believing that I think 14kbps is transparent. I don't have to do one but by doing an ABX and not caring by just pressing randomly the result will back up my claim that 14kbps is transparent. But if I claim that 14kbps is not transparent I am enforced to do an ABX. This is weird.

Actually, shouldn't the claim transparent be a TOS #8 since we can not easily prove it? We can prove though that we can hear a difference by getting good results on an ABX and therefore the only thing we can prove is that we do hear a difference and not when it comes to not hearing a difference.

Please let me know if I am missing some crucial scientific evidence of the opposite or anything else. Regards.

Re: Opus 1.3-beta is here

Reply #24
Anyway, the track is full of artifacts which makes me wonder what age you are and if you have suffered any ear damage. I promise, I am not at all being rude but I felt I had to ask since the quality is bad and you say it is transparent to you. Regards.

I winced when I saw him claim transparency at 48kbps as well because to me just about any full bandwidth music is obviously degraded at that bitrate. But, I can play 48kbps mp3 files for my wife and kids and they just can't hear the artifacts. I don't know what it is but they swear the files sound just fine to them. To me they sound like someone drowning kittens.

I've come to believe that some people just don't have the capacity to detect/notice/care about audio faults and degradation. I'm kinda jealous because they can store 5 times the amount of music on their devices since it can be encoded at stupidly low rates. They can also get by perfectly fine on $5 garbage can quality headphones and think they are getting perfect fidelity.

I'm in my mid-40's and played bass in a metal band most of my life, so my hearing is far from what anyone would call good. That being said, I'm the only one in my family that ever notices and complains when audio plays which is loaded with artifacts (it's really bad on Sling TV for instance). For me, Opus tends to get near transparency around 96kbps. Around 112kbps it gets to be essentially impossible for me to ABX. I know there are people who can, but that is about the level my ears and brain tap out.

 
SimplePortal 1.0.0 RC1 © 2008-2018