Hacker News Reader

Comment by jchw | Hacker News Reader

It seems like there's an answer: the market actually wants free standards because they're realizing it's more beneficial to them than any other outcome. So money and effort is being pumped into developing free video encoding standards. It seems there is an incentive for the foreseeable future to continue to improve video codecs simply because it saves on bandwidth and disk space costs, and improves user experience.

Though obviously, everything's too good to be true. Just like MPEG seems to be hitting the end, some day the AOM will hit the end, too. The world changes. Something else will emerge and impact the business model. It may be worse, or maybe it'll be a return to norms. Could just be another one of those things that cycles every so often.

With audio codecs, it doesn't look like we ever really needed much in the way of commercial entities, since Vorbis and Opus have been outperforming the competition for a long time now.

View on HN

> …Vorbis and Opus have been outperforming the competition for a long time now.

Sorry to be pedantic about one bit of an otherwise-fine post, but AAC is comparable to Opus at any given bitrate[1]. Plus, even the most esoteric flavor of AAC has the enormous advantage of 7+ years of ubiquity[2].

[1] See the multiple charts at http://opus-codec.org/comparison/, for one presumably-unbiased source.

[2] All versions of Android support HE-AACv2 playback, iOS introduced support for HE-AACv2 playback in iOS 4 (released 2010), macOS introduced support for HE-AACv2 playback with iTunes 9.2 (released 2010), Adobe Flash Player introduced support for HE-AACv2 in 2007, All versions of Windows Phone support HE-AACv2 playback, all open source players support HE-AACv2 playback via FAAD2

OK, fair enough: HE-AAC had been out-performing the leading open source audio codec for a while. Easy to forget primarily because even after many platforms supported HE-AAC, it wasn't necessarily common across all media. I really saw it explode on the internet around the time DASH streaming became more common, since adding new audio formats was much less expensive.

That being said, I still think Opus is better now, since it is more versatile. I've yet to see anyone use HE-AAC in a real-time context, unless they're using it under a different name.

That comparison appears to says that the difference in quality between opus and he-aac is the same as the difference between he-aac and vorbis.

Opus is definitely better, but we have basically hit the wall in audio compression (we'll need a serious breakthrough to go much farther). There also isn't as much interest in audio codec research because even very large raw audio files pale in comparison to fully compressed video.

Incremental improvements are happening in opusenc, current status (as of 1.2 and 1.3-beta) is that musical content is generally transparent at ~96kbps VBR, with a few needing ~128kbps and a tiny minority needing ~160kbps.

The developers are slowly reducing the required bitrate for transparency, it wouldn't surprise me if they reach transparency at ~64kbps VBR in a year or two.

Can you provide links to blind tests supporting these conclusions?

https://www.opus-codec.org/comparison/

https://i.imgur.com/zxwOBMJ.png

Etc.

Discussion of 1.3-beta here:

https://hydrogenaud.io/index.php/topic,115156.0.html

Discussion of preferred bitrate and point of transparency:

https://hydrogenaud.io/index.php/topic,115386.0.html

https://hydrogenaud.io/index.php/topic,114656.0.html

I would highly recommend that you give it a go yourself. Take a FLAC source, encode Opus files yourself. The encoder is freely available from https://opus-codec.org/downloads/ and integrates nicely into Foobar2000's converter function (Opus is already there as a profile, it will ask for the location of opusenc.exe the first time you use it).

Use the ABX Comparator plugin for Foobar2000 (both available from http://foobar2000.org) and run a few trials.

After all, public listening tests are one thing, but to be absolutely sure a format/bitrate is transparent to you, individual testing is needed.

I have run a number of tests myself (of Opus and other formats) and found general transparency at ~96kbps Opus, lower for some tracks. For those tracks that weren't transparent at that bitrate, the artifacts were generally benign and not that annoying (slightly increased noise levels, mostly).

Oh god, let define a few thing.

Generally Transparent = MP3 LAME @ ~ 128kbps. That is the good old standard.

There has been PR and marketing materials, much like Video Codec making claiming they are 50% better then H.264 or AVC or whatever. AAC or MP3Pro even claim to be MP3 128kbps quality at 64Kbps. Of course which never happened even after years of fine tuning.

Opus is the first and only codec in years, or decades that had better music quality then MP3 @ 96Kbps. Even AAC cant do that, at least not in majority of cases.

I have come to the conclusion that Audio compression has come to end of the S curve, with diminishing returns. It literally took us all the years till all MP3 patents has expired to get an Audio codec that is better at it with 20% less bit rate.

I very much doubt we could have 64Kbps VBR to sound better without many more breakthroughs.

>Generally Transparent = MP3 LAME @ ~ 128kbps. That is the good old standard.

Says who? And which version of LAME? That encoder has seen huge improvements over the years. We should not define audio transparency by an outdated and flawed format. We should define audio transparency by whether it is distinguishable from a lossless source or not.

But yes, a modern version of lame (3.98 or newer) will be transparent around V5 VBR, which is ~130kbps. BUT -- and this is a huge twerking 'but' -- the MP3 format suffers from pre-echo. There is nothing you can do about it. It is lessened at higher bitrates, but you can never completely get rid of it. There is also the problem of the SFB21 defect, which can severely bloat the allocated bitrate on content that is heavy in high frequencies: http://wiki.hydrogenaud.io/index.php?title=LAME_Y_switch

Newer formats such as AAC, Vorbis and Opus do not share these limitations. I don't know where you read that AAC is worse than MP3, because that is simply objectively false, as shown by every listening test you can care to dig up. Did you mistakenly use FAAC (probably the worst AAC encoder out there) or something?

I never said AAC is worst, but AAC 96Kbps cant beat MP3 128Kbps in majority of cases.

Yes it can. It absolutely can, and does.

https://i.imgur.com/zxwOBMJ.png

I know it's one listening test out of many, but they all agree. AAC (with a good encoder, such as Apple's) at ~96kbps is better than MP3 at ~130kbps.

HE-AAC has the colossal downside that there is no free software encoder.

FFmpeg/libav don't do HE. FDK/libaacplus are non-free. neroAacEnc always used to have the best quality, but that's shareware.

> See the multiple charts at http://opus-codec.org/comparison/, for one presumably-unbiased source.

Further nitpick: the page is very misleading. Audio compression has different targets, a crucial one being music.

The Opus official page has comparatively little test references for music, both of which are at 64/96kbps for AAC [1-4], which is significantly below the general audience listening standards (160+ kpbs).

This implies that the red line on the top right of the first diagram is literally made up.

Until further blind listening tests are performed, generic conclusions such as "Opus outpeforms the competition" are unfounded.

> It attempts to summarize results [...] and (when no data exists) show anecdotal evidence

This is borderline offensive for people who are serious about researching music quality. Any anectodal evidence [in this field] is garbage.

[1] http://listening-tests.hydrogenaud.io/igorc/results.html [2] https://people.xiph.org/~greg/opus/ha2011/ [3] http://listening-test.coresv.net/results.htm [4] http://opus-codec.org/static/comparison/GoogleTest1.pdf

There's a good reason all the listening tests have stopped at 96 kb/s. Above that, the quality of Opus, Vorbis and AAC is so close to transparency that it's pretty much impossible to get statistically significant results. Even the latest 96 kb/s test was really stretching things.

Reply on news.ycombinator.com