Hello!

I already learned that a step from -q4.99 to -q5 results in a significant quality increase because of different stereo algorithm. Also the bit-rate jumps.

My question is about settings q3.99 and q4. Does the quality jump in this case?

In my tests I can't hear much difference actually, and the bitrate increases quite smoothly from -q3 to -q4. But maybe there is some hidden algorithmic change that favours -q4 over -q3.99 in general?

Regards,

/Konstantin [new to ogg-vorbis]

P.S. I want something about 120-128 kbps -- only for background listening and maybe for a portable. And have found -q3.5 gives quite acceptable bitrate.

No.

Technically, the codebooks are switched at each (x-1).99 -> x.0 level switch, but that will not be audible.

--
GCP

Basically the switch between -q 4.99 and 5.00 like Garf stated performs more quantization in the codebooks, anything below 4.99 is coupled and quantized in a lossy form (point stereo localization) above 5.00 are coupled losslessly, Basically the coodebooks that are used in this case are Vector Quantization Lattice / Huffman Codebooks which are static for the time being.

This is not what I said.

What I said is that the (fixed, predetermined at encode time) codebooks get switched at [b]each x.99 -> (x+1).0 change. But this will not imply a big quality/filesize difference.

Additionally, the switch between 4.99 and 5.0 switches on lossy stereo coupling, which will make a big difference.

Lossless quantization sounds like an oxymoron to me.

Edit: To summarize, if you have to pick between -q3.99 and -q4.0, pick 4.0

--
GCP

A) fixed the oxymoron had't realized that before
B) is there any real straight answer to the exact codebooks that used? you say they get switched, but what are we talking about vector quantization or huffman codebooks? I asked this before and nobody could figure this out. I am not pushing the matter I just want some idea. This is why I oppose the idea of having to use the -q switch all of the time, especially if you are experimenting or if you are a developer. Surely it's made for it's simplicit ease of use, but in RC4 when Monty finishes working on I would really like to be able to choose my own free average bitstream like in the previous version (RC2) with some optmizations to the physcoacoutics model for using that switch. I am sure Monty is handling this though.

Both get switched.



If you are a developer, you change the codebooks used in the code.



This is a complete misunderstanding. You have _never_ set the average bitrate. You have always selected a -q level, even with RC2. The tools presented it as an average bitrate, but in reality you have always been choosing a quality level. Because the RC2 encoder was quite stable, a given -q level corresponded quite well with a certain bitrate, which is why it was presented as such.

The Vorbis codec is natively VBR and quality based. Setting an average bitrate and ' applying optimizations to the psychoacoustic model for that switch ' make no sense. The psymodel _determines_ the bitrate.

--
GCP

Garf, thanks for your reply!



ok, I guess now the choice is between anything from q3.5 to q4, as you've confirmed that quality improves gradually with the bit-rate. So it's a plain trade-off between slightly lower bit-rate and slightly higher quality. Gotta make more tests on my own.

Thanks again.

/konstantin

Sorry, I had been referring to the managed bitstream switch, sorry for the misunderstanding. developers change the codebooks you mean the subsets that are built in backends.h? I couldn't find that it never came with the source-code.