2pass?

[zvezdochiot]

2- and more- pass coding, like nero?

[rillian]

I don't know of anyone working on 2-pass rate control for Vorbis.

Can you share a use case for this? Usually two-pass is used to maximize quality within a strictly limited number of bytes, but vorbis is transparent at any current media readout or broadband streaming bitrate.

[zvezdochiot]

It is not so easy to answer your question.

Additional regulators (multipliers) are introduced into the coding system, initially equal to 1. Depending on where they are entered, the corresponding statistics are collected on the first pass. There is also a flag system that determines which regulatorswill be enabled. Based on statistics, the values of the regulators are calculated. In general, this system is empirical in nature and cannot be fully described using a certain theory.

PS: For stability, the values of the regulators are usually calculated by delta-functions:

regulator =  (freg) ? (1.0f + min(abs(f(stat)),deltaMax)*sign(f(stat))) : 1.0f;

[zvezdochiot]

:information_source: The general meaning of multi-pass coding: to partially transfer control of the coding process to the sound itself (feedback).

PS: Multi-pass is no magic.

[rillian]

I understand what 2-pass encoding is. I do not understand why you want it for vorbis. Can you explain why it would be good?

[zvezdochiot]

Encoding, in which nothing is known about the encoded signal, in no way can have little bad. Only the encoding of a known signal (statistics accumulated) can have little bad.

PS: little bad != best || good

[rillian]

Well yes, but as you say, multi-pass encoding is not magic. It doesn't make every encoding better.

Two passes are only helpful to adjust the bitrate for different sections to meet some external constraint. That's typically only useful for video because video data is so large, so one might want either to avoid overwhelming someone's network bandwidth, or to exactly fill some distribution medium like a video disk. Those constraints mean video is often compressed so heavily that it's visible. You really are trying to spread around the bad so it's less noticeable.

What the current vorbis encoder does instead is just to make the file a little bigger for every piece that is harder to encode. This works fine since audio is so much smaller. Why doesn't that work for you?

[zvezdochiot]

This "good" only works at high bitrates.