Opus 1.3 Released - One Of The Leading Lossy Open-Source Audio Codecs

[Weasel]

But it in fact is the job of lossy audio compression to discard inaudible data, in particular, unpredictable inaudible data; lossy compressors don't mind coincidentally encoding inaudible data if those data are produced by the predictor at no additional cost. Discarding unpredictable inaudible data is the process by which a lossy audio compressor "compress[es]", and rate control is how it does so "at a given bitrate on average."

I already answered this and it's simply wrong.

The job of lossy compressor is not to DISCARD data, but to encode at a given bitrate. You can discard inaudible data WITHOUT compressing anything so does it mean that would also be a lossy compressor? Really?

Its quality depends on how it achieves that, and yes usually it discards inaudible data for quality, but that is to fulfill its goal, it is NOT THE GOAL, it's a means to an end.

Not really sure what you're getting at here.

Seriously? Let's imagine a fictional scenario for simplicity.

We have two codecs: Foo and Bar. We compress some signal, both at 256kbps. Their job is to discard inaudible data, right?

Foo discards all inaudible data.
Bar discards less inaudible data, because it has less redundancy in its algorithms and encodings -- it's simply "better" at encoding the same information.

Which codec is better?

According to YOU, Foo is superior, since well, the job is to discard inaudible data, and Foo does that and discards MORE DATA than Bar. Which is obviously wrong.

Now replace Foo with mp3, and Bar with AAC and we suddenly turned this into reality.

Tell me again how mp3 is superior to AAC because, obviously, discarding data is the goal of a lossy codec so mp3 is simply the best lossy codec!

Almost all MP3 encoders were capable of encoding wideband audio, and that was already the default width for standard music bitrates by the time anyone was really using it. MP3 is missing proper tools for encoding frequencies above 16kHz, so MP3s with any frequency content above 16kHz are basically a novelty, even if LAME manages to start to do something useful in that range at outrageous (180kbps+!) rates. To me, it seems like basically all MP3 encoders did wideband encodes then, and do little more than wideband encodes now.

I don't know what world you live in, but 320kbps mp3 are the norm these days, and you call 180kbps outrageous? Wat? Anyway, it has nothing to do with what I said, see above, hopefully you get it.

At about 32kbps, Opus will enable fullband encoding; so virtually any music-rate Opus encoding will have more bandwidth than virtually any standard rate MP3.

But that makes it worse according to you, since it discards less. Isn't the job of a codec to discard data?

So Opus is worse since it discards less bandwidth.

Your logic, not mine.

At a given bitrate, chances are that MP3 will be wideband (or somewhere between WB and FB), and Opus will be fullband. That is to say "These days", "[MP3 encoders]" don't "tend to cut as much as Opus but not more", but rather they tend to cut more than Opus at almost any given bitrate. The reason for this is simple: MP3 makes encoding frequencies above 16kHz very expensive, because it barely even designed with this in mind; if some of the people at MPEG had their way, they would have specified the decode sample rate at 32kHz and washed their hands of the whole ordeal.

Yes, you are right, but funnily enough you missed my point completely. (again, see above)

Sorry, that was a bit snooty of me, and I'm no expert, but it just seems weird to me to snuff out somebody's good first instinct "compression is about inaudible data" with a hard denial, and then go on to explain why it's a good instinct.

No, I never said mp3 is better than Opus, I was using your logic to prove why it's nonsense to say that the job is to discard.

I did say, however, that AAC is better than Opus, at least at high bitrates (i.e. 256kbps or more). Opus is obviously king at low bitrates, but I don't care about that.


tl;dr: the job of a lossy codec is to encode at a given bitrate. How it achieves that (by discarding inaudible data) is NOT the job or goal, it's simply a means to an end.

[Brisse]

Weasel Please explain what kind of use-case we are talking about here. When is lossy compression at extreme high bitrates useful? In production one obviously avoids lossy at all costs as that would more or less be a crime, and even as an end user or consumer there comes a point where one has to consider lossless compression instead of lossy @ high bitrates. I feel like a lot of people use codecs the wrong way. Use the right tool for the job. If you are going to hammer in a nail, use a hammer, not a screwdriver. Sure, 320kbps mp3's might be some sort of norm, but it's stupid and shouldn't be encouraged. It's an old outdated codec at a bitrate that doesn't make sense.

[Tomin]

I already answered this and it's simply wrong.
The job of lossy compressor is not to DISCARD data, but to encode at a given bitrate. You can discard inaudible data WITHOUT compressing anything so does it mean that would also be a lossy compressor? Really?

Its quality depends on how it achieves that, and yes usually it discards inaudible data for quality, but that is to fulfill its goal, it is NOT THE GOAL, it's a means to an end.

While true, I think the point was that the measure of goodness here is not how much data the compressor can conserve but what is the perceived quality. These are audio codecs and not general purpose compressors. For humans it doesn't matter if the codec will conserve higher frequencies because they are inaudible anyway so they do not affect the perceived quality. Of course if the compressor filters frequencies too low then it affects.

I hope I used the correct words because this is far away from my area of expertise and English is not my first language.

[microcode]

The job of lossy compressor is not to DISCARD data, but to encode at a given bitrate. You can discard inaudible data WITHOUT compressing anything so does it mean that would also be a lossy compressor? Really?

Yes, it does.

[Weasel]

Yes, it does.

No it doesn't.

Maybe in your world compressor means "spit out the same file size but remove information".

[Weasel]

Weasel Please explain what kind of use-case we are talking about here. When is lossy compression at extreme high bitrates useful? In production one obviously avoids lossy at all costs as that would more or less be a crime, and even as an end user or consumer there comes a point where one has to consider lossless compression instead of lossy @ high bitrates. I feel like a lot of people use codecs the wrong way. Use the right tool for the job. If you are going to hammer in a nail, use a hammer, not a screwdriver. Sure, 320kbps mp3's might be some sort of norm, but it's stupid and shouldn't be encouraged. It's an old outdated codec at a bitrate that doesn't make sense.

Yes, I despise mp3 also, it's not what I meant there at all, since I was fully sarcastic with my mp3 comments (and why his logic is absurd).

But 256kbps AAC is the standard for high quality music. Lossless gives almost no benefits for it and it's easily 2-3 times larger (depending on content). Of course, lossless is good for archival if you plan to transcode it in the future to something else. You can do a diff (subtract one audio from the other) with 256kbps AAC and lossless (take care to shift the lossless one since AAC adds a silence block at the beginning). You'll end up with something extremely faint which is not worth caring about.

But anyway, the point was that someone asked what's better than Opus.

And AAC is better at high bitrates; it drops less information for the same bitrate.

[Weasel]

While true, I think the point was that the measure of goodness here is not how much data the compressor can conserve but what is the perceived quality. These are audio codecs and not general purpose compressors. For humans it doesn't matter if the codec will conserve higher frequencies because they are inaudible anyway so they do not affect the perceived quality. Of course if the compressor filters frequencies too low then it affects.

I hope I used the correct words because this is far away from my area of expertise and English is not my first language.

"Perceived quality" is quite subjective and thus I wasn't talking about it at all.

Instead, if you have two codecs (i.e. mp3 and opus) and one cuts everything above 18k (mp3) while Opus keeps up to 20k, and they have the same bitrate (256kbps), then clearly Opus is superior since it discards less information. And it's the same from Opus to AAC at such bitrates. AAC discards less, so it's simply superior.

Of course, we're talking high quality encoders here, so they know when to start encoding the high freqs, that is when you can't really improve the quality of the lower spectrum anymore.

I don't care if you (or anyone else, or even me) doesn't hear those frequencies that AAC keeps. That's irrelevant. It's still superior.

[microcode]

"Perceived quality" is quite subjective and thus I wasn't talking about it at all.

Instead, if you have two codecs (i.e. mp3 and opus) and one cuts everything above 18k (mp3) while Opus keeps up to 20k, and they have the same bitrate (256kbps), then clearly Opus is superior since it discards less information. And it's the same from Opus to AAC at such bitrates. AAC discards less, so it's simply superior.

No, it's superior because it predicts more useful information, and discards more useless information. Even if you low pass filter the signal before handing it to either codec (or the signal lacks frequencies in that range to begin with), Opus will come out sounding more accurate. Furthermore some aspects of "perceived quality" are not subjective, but physiological and similar among most humans*, such as nonlinear sensitivity to frequencies, auditory masking, or implied missing fundamentals. These physiological and psychological effects are the primary source of improvements in coding efficiency in new codecs.

If psychoacoustics were entirely subjective, and thus non-uniformly objective, we would probably just be transmitting PCM at different rates.

You keep saying these things, and it's clear that you know a handful of things, but you say more than you know, and do so with so much certainty. If you think that the primary qualitative difference between MP3 and Opus is bandwidth, then you are utterly mistaken, and should not be lecturing people on this topic.

* Physiological effects are often lumped in with psychological effects (also measurable and common to humanity) to form "psychoacoustics".

[Weasel]

No, it's superior because it predicts more useful information, and discards more useless information. Even if you low pass filter the signal before handing it to either codec (or the signal lacks frequencies in that range to begin with), Opus will come out sounding more accurate. Furthermore some aspects of "perceived quality" are not subjective, but physiological and similar among most humans*, such as nonlinear sensitivity to frequencies, auditory masking, or implied missing fundamentals. These physiological and psychological effects are the primary source of improvements in coding efficiency in new codecs.

If psychoacoustics were entirely subjective, and thus non-uniformly objective, we would probably just be transmitting PCM at different rates.

You keep saying these things, and it's clear that you know a handful of things, but you say more than you know, and do so with so much certainty. If you think that the primary qualitative difference between MP3 and Opus is bandwidth, then you are utterly mistaken, and should not be lecturing people on this topic.

* Physiological effects are often lumped in with psychological effects (also measurable and common to humanity) to form "psychoacoustics".

Most of the time the bandwidth depends on the encoder, not just on the codec. Opus is an exception that hard caps the audio at 20k.

It's simple logic, really. You don't need any expertise to prove this. If an encoder decides that it's worth it to encode frequencies above 20k, it must mean it has done a perfect job (acoustically) on lower frequencies. Why else would it start encoding freqs above 20k in AAC?

Which means that AAC is simply superior, as you get "inaudible frequencies" on top of all the audible spectrum. You get more info, even if you find it useless, so the codec is superior.

Alternative is... everyone who wrote any encoder is a moron, except for Opus. Makes total sense, not. But you can continue to believe in this delusion.

[SciK]

It's simple logic, really. You don't need any expertise to prove this. If an encoder decides that it's worth it to encode frequencies above 20k, it must mean it has done a perfect job (acoustically) on lower frequencies. Why else would it start encoding freqs above 20k in AAC?

Such trust is beautiful, but it seems a little naïve.

Let me reiterate my previous question: if we assume that it has indeed done a perfect job on audible frequencies, why bother at all with the higher frequencies and create a bigger file as a result?