44 or 48khz? and ogg bitrate

[Apfelstruhdl]

i use ogg for all my movies now and was wondering if vorbis gives better quality at same bitrate with 48 or 44khz? and how big is the quality loss when downsampling?

Which progie do you use when creating oggs from ac3? i tried oggmachine several times and heard about headac3 which i think gets much tweaking by darkavenger. so which do you prefer?

last question: which settings for ogg are you using? I tried -q 2 several times and was pretty impressed by the quality. It gives around (96kb). can you even use lower bitrates for one cd rips? i don't know so much listening tests so give me your opinion :D .

[tangent]

There are only two reasons to downsample to 44.1kHz

1. Your soundcard cannot handle 44.1kHz
2. You intend to encode with LAME MP3, which is known to have quality issues with 48kHz

I choose to use HeadAc3 for Ogg transcoding because I am comfortable with the 1-pass normalising, which is faster but you need some HD space for it.

I choose to use -q0 for my encodes. I know if I listen carefully I can easily distinguish between the original and -q0, but Ogg Vorbis provides -q0 where the problems are not annoying, and usually not perceptible while watching video.

[DSPguru]

@tangent
you know this is NOT a one-pass normalization. the pcm samples are processed twice -
1. one pass for finding maximum gain
2. second pass to deliever the samples to the encoder.

the only program that has one-pass normalization is BeSweet, where the search for maximal gain is done WHILE encoding to MP3.
as you probably know, only MP3 headers have global_gain property that allows implementation of post-gain.

[tangent]

I must have the wrong idea of what 1-pass normalization means. What I meant was I use HeadAC3 because (what I believe is happening from my observations) it decodes AC3 to WAV in 1-pass and determines the gain, and applies this gain while encoding from WAV to Vorbis. If you remember correctly, you told me before that this feature is not available in BeSweet.

Of course, I would prefer to use BeSweet over HeadAc3 when encoding to MP3. But for encoding to Vorbis, I prefer HeadAc3.

[MaTTeR]

@tangent

Using q0 in HeadAc3he, do you leave the Channel Coupling at Heavy Lossy or lock it in to something else? I'm curious how it affects the overall quality at q0.

[DSPguru]

ok, so now we agree on the facts :).
no, it isn't available. BeSweet was invented as a reply to a thread (http://forum.doom9.org/showthread.php?s=&threadid=7245) of Boost who asked for a way to transcode without intermediate file. BeSweet solved the problem of a huge space needed for audio transcoding :D.

Dg.

[tangent]

Originally posted by MaTTeR
@tangent

Using q0 in HeadAc3he, do you leave the Channel Coupling at Heavy Lossy or lock it in to something else? I'm curious how it affects the overall quality at q0.
I leave it as heavy lossy. I have no idea of how changing to lossless or less lossy will affect the quality because I haven't tried. But it is in my opinion that if you want to spend more bits, it would be wiser to just increase the -q level.

[DarkAvenger]

To be most precise something like "one pass normalization" doesn't exist.

About ch coupling: Maybe for better DPL, you should choose light lossy or even lossless.

[tangent]

I really hope to get a DPL setup one day so I can actually test all these things... but I was thinking, shouldn't it be easy enough to simulate the setup? Decode to wav, extract the rear channel, and test against the original rear channel...

[Peters]

Originally posted by tangent
There are only two reasons to downsample to 44.1kHz

1. Your soundcard cannot handle 44.1kHz
2. You intend to encode with LAME MP3, which is known to have quality issues with 48kHz

I choose to use HeadAc3 for Ogg transcoding because I am comfortable with the 1-pass normalising, which is faster but you need some HD space for it.

I choose to use -q0 for my encodes. I know if I listen carefully I can easily distinguish between the original and -q0, but Ogg Vorbis provides -q0 where the problems are not annoying, and usually not perceptible while watching video.

With such a low bitrate, downsampling to 44.1 khz is better. See John33 amswer in Hydrogenaudio forum

http://www.hydrogenaudio.org/forums/showthread.php?threadid=1015&highlight=downsampling

[DSPguru]

@tangent
writing a DPL decoder isn't that trivial.
the only thing that is well-defined, is the DPL encoder. as for decoding DPL, every1 could create its own algorithm that adapts the reconstruction matrix.

[MaTTeR]

@Peters

I suspect that downsampling to 44.1kHz may no longer be required with the upcoming RC4. Tangent could speak more to this but as I understand it, lower bitrate tuning has been a large focus of RC4.

[tangent]

Originally posted by Peters


With such a low bitrate, downsampling to 44.1 khz is better. See John33 amswer in Hydrogenaudio forum

http://www.hydrogenaudio.org/forums/showthread.php?threadid=1015&highlight=downsampling

I asked Monty in IRC a few months back.

[14:13:53] <tangent3> xiphmont: is resampling to 44.1kHz recommended for low low bitrates?
[14:14:07] <xiphmont> No, why would it be?
[14:15:26] <xiphmont> what resampling gets you is greater effective frequency discrimination in the encoder, because you can use the same block size (for long blocks at least) on a longer time segment. but 44.1 is not magic, and not worth it from 48kHz.
[14:16:47] <xiphmont> 48->32 or 22 or lower (for really low bitrate) makes sense.

[Peters]

@tangent

Thank's for the feedback

Well, it seems that the question of downsampling or not don't make unanimity :(

John and xiphmont agree with the better 48->32 downsampling but John said that ogg is not tuned for anything other than 44 or 48.

So, all is not clear for the moment. Maybe RC4 will make automatic downsampling like lames does?

[tangent]

I would prefer to trust Monty's advice on this matter. I hope you know who he is.

[Peters]

Originally posted by tangent
I would prefer to trust Monty's advice on this matter. I hope you know who he is.

:) yes i know the ogg project team
By the way i'm sure you know who is John33

So it's very strange to have two different points of view, don't you think?

[tangent]

Well, John33 compiles all the source code which Monty writes.