HydrogenAudio

Lossy Audio Compression => Ogg Vorbis => Ogg Vorbis - Tech => Topic started by: Jon Ingram on 2001-12-22 14:42:48

Title: Vorbis RC3 quality settings
Post by: Jon Ingram on 2001-12-22 14:42:48
RC3 is almost ready to do, and now has two ways to specify encoding:
1) -m minimum -M -maximum b  nominal
2) -q quality

The quality indication goes from 0, which is nominally 64 kpbs, to 10, which is nominally 480kpbs, and is a floating point number.

To get some indication of how bitrate changes as you change the quality setting, take a look here:

http://www.maths.warwick.ac.uk/~jingram/ogg/ (http://www.maths.warwick.ac.uk/~jingram/ogg/)

I (along with xercist from the #vorbis IRC channel) graphed bitrate-vs-quality for several well known samples, including hard-to-encode ones such as applaud.wav and fatboy.wav. The overall graph is as follows:

http://www.maths.warwick.ac.uk/~jingram/ogg/group.png (http://www.maths.warwick.ac.uk/~jingram/ogg/group.png)

The big jump at q=5 is mainly down to the switching to lossless channel coupling. You'll also notice that the bitrate for q=n.99 is generally *larger* than that for q=n+1-- this is because the same settings are used for q in [n, n+1), and they are tuned for the  lower end of that range.

Have fun.
Title: Vorbis RC3 quality settings
Post by: john33 on 2001-12-22 15:11:14
The daily binary at rjamorim's site, has the quality option enabled and works fine. As examples, -q 5 produces an ABR of about 160kbps and -q 6 about 192kbps.

john33
Title: Vorbis RC3 quality settings
Post by: karajan on 2002-01-05 05:48:18
Just for reference, I measured cut-off frequency(of LPF) at quality setting, by using white noise.
Also, it seems like white noise gives rise to much more bit rate than actual music.

http://www2.ocn.ne.jp/~mp3lab/exp_lab/vorb...vorbis_rc3.html (http://www2.ocn.ne.jp/~mp3lab/exp_lab/vorbis_rc3/vorbis_rc3.html)
Title: Vorbis RC3 quality settings
Post by: Jon Ingram on 2002-01-06 03:34:58
I've updated all the graphs to RC3 (no actual changed in bitrate, given that the encoder on 22/12 basically *was* RC3). I've also added a couple more hard to encode test samples.

In particular, I have a white noise sample. Interestingly, for qualities below 5, encoding this does not expend an enormous amount of bits -- but this soon changes for qualities above 5.5.