The new Opus encoder is incredible. We used the beta of 1.3 in a bandwidth constrained environment recently and found that in our testing (and for our very specific set of audio, YMMV, past performance not indicative of future blah blah blah), nobody could reliably distinguish between an AAC at 128kbps and an Opus file at 64kbps. If we tried really hard, we thought we could identify some extra aliasing in the top end, but this was not practically relevant and never picked up by our user testing. The audio in question was a mixture of speech and quiet music; of course the different encoders were easily identified when using regular high-density pop music!
The files were fully supported by every Android device we had available for testing.
Yeah, I've been testing 1.3 and my experience lines up.
All of the comparisons are amazing, but I'm most impressed by the SILK improvements at 9kb/s. The sample is just barely noticeably worse than the reference with 1.3, and artifact-ridden with 1.2 and prior.
Opus 1.3 at 9kb/s and above is perfectly readable, and unnoticeable aside from some sibilants for voice, that's spectacular.
I remember Jean-Marc Valin, the guy from Xiph published a noise suppression algorithm using RNN a year ago [1]. I thought it was just an experimental project that will be abandoned soon. But they really integrated RNN to the codec.
This is really amazing and deserves more attention. Audio codec work has been kind of neglected lately, with the increases in bandwidth and storage making near-transparent lame v0 mp3s the obvious choice for many applications, but almost unnoticed except to codec nerds like myself, Opus kept improving and was already head and shoulders above any other codec. That includes Vorbis and AAC.
We're now at the stage where 96 kb/s stereo audio is reliably transparent, with the occasional quirky sample that might require 128 kb/s or 140 kb/s for full transparency. That's a remarkable achievement. I applaud the Opus team for their hard work.
I was thinking of "alien technology from the future" (an expression incidentally coined by Xiph's Tim Terriberry) when I read that they somehow managed to improve Opus.
Technically, the linked article is correct that it is something that I like to say, but I stole the phrase from David Schleef (one of the Dirac developers).
If the reports are true, sounds like they should just have been bold and called it Opus 2. Sounds like true next gen codec performance. I need to get hold of some FLAC’s now and have a listen...
The reason we are not calling it Opus 2 is that it could confuse some people into thinking we broke compatibility. Opus 1.3 is perfectly compatible with Opus 1.0, and all future releases will keep that compatibility.
Like many other audio codecs, Opus lets the encoder decide how to spend the bits is has -- on what frame and on what frequency bands. On top of that is has a few special features that also require decisions from the encoder. So while the decoder doesn't change, the encoder can be improved to make better decisions. While the format itself is not perfect, I have not come across any particular thing that would be worth breaking compatibility over. I prefer working within the constraints of the bitstream to keep improving the quality.
I'm always amazed by how Opus manages to utterly crush each one of its competing codecs. I'm also amazed by the fact that finally we have a widely supported free codec no MPEG-LA covered codec can compete with. Combined with AV1 and Matroska, this is an amazing victory open formats.
The new Opus encoder is incredible. We used the beta of 1.3 in a bandwidth constrained environment recently and found that in our testing (and for our very specific set of audio, YMMV, past performance not indicative of future blah blah blah), nobody could reliably distinguish between an AAC at 128kbps and an Opus file at 64kbps. If we tried really hard, we thought we could identify some extra aliasing in the top end, but this was not practically relevant and never picked up by our user testing. The audio in question was a mixture of speech and quiet music; of course the different encoders were easily identified when using regular high-density pop music!
The files were fully supported by every Android device we had available for testing.
Very impressive work, thank you Xiph!
All of the comparisons are amazing, but I'm most impressed by the SILK improvements at 9kb/s. The sample is just barely noticeably worse than the reference with 1.3, and artifact-ridden with 1.2 and prior.
Opus 1.3 at 9kb/s and above is perfectly readable, and unnoticeable aside from some sibilants for voice, that's spectacular.
Great job Xiph!
[1]: https://news.ycombinator.com/item?id=15362982
We're now at the stage where 96 kb/s stereo audio is reliably transparent, with the occasional quirky sample that might require 128 kb/s or 140 kb/s for full transparency. That's a remarkable achievement. I applaud the Opus team for their hard work.
> Any sufficiently advanced technology is indistinguishable from magic.
https://people.xiph.org/~xiphmont/demo/daala/update1.shtml
Good read: http://dafoster.net/articles/2015/03/14/semantic-versioning-...
If you could break compatibility, what would you change and why?
Clicking on the "xiph.org" image at the top directs to a 404.