Best Conversion method?

[mstic]

I'm decoding an ac3 track with AC3DEC to a 48khz PCM wave

Whats best way to convert the 48khz to a 44.1 stream?

using ac3dec - 44100
or possibly a different encoder?

[tangent]

http://66.96.216.160/cgi-bin/YaBB.pl?board=c&action=display&num=1010178496

[ChristianHJW]

SSRC ..... best used in BeSweet in 32 bit floating point mode ....

[Rumata]

Originally posted by mstic
I'm decoding an ac3 track with AC3DEC to a 48khz PCM wave

Whats best way to convert the 48khz to a 44.1 stream?

using ac3dec - 44100
or possibly a different encoder?
Why to convert it to 44.1 KHz ?

[MaTTeR]

I personally don't see any reason to downsample a file to 44khz unless you have a really old and nasty soundcard. Can someone else enlighten me why the need to downsample?

BTW- Use SSRC if you do convert.

[LotionBoy]

someone mentioned that Lame has problems with 48kHZ files. That could be a reason to downsample. I haven't noticed any, but I don't have the best ears in the world.

LotionBoy

[tangent]

LAME problems with 48kHz encoding discussed in this thread:

http://www.hydrogenaudio.org/forums/showthread.php?s=&threadid=275&highlight=48khz

http://www.hydrogenaudio.org/forums/showthread.php?s=&threadid=432&highlight=48khz

[ChristianHJW]

No ..... i won't reply here :D :cool: :D

[MaTTeR]

@tangent
Thanks for the links. It appears as though Dibrom prefers 44khz because 48khz might cause pre-echo. My ears have never heard this artifact but it's nice to know it could exist.

@ChristianHJW
Been there and done that:D I remember the last time this subject came up and got sort of nasty.

[Doom9]

did dibrom ever submit any evidence for his claims? I've been doing 48khz for a long time and never had any problems.. but then again I'm more into visual quality than audio quality and my audio equipment isn't really high end

[MaTTeR]

I play all my encodes on a high end entertainment system and can certainly say I've heard no artifacts @ 48khz. I'm very picky with both audio & video quality but perhaps my ears just don't hear this pre-echo.

AFAIK Dibrom has just stated he heard some of the artifacts on a few samples but not all. I haven't seen any graphs showing what he's talking about though.

[DarkAvenger]

I don't recommend training your ears to hear artifacts. So will start hearing to much flaws. :) (No, I am not belonging to the trained persons, though I can hear whether music is lousily encoded at 192kbps or if it is pcm.)

[tangent]

Graphs say nothing about audio encoding quality. Nothing beats the ear. Graphs can be used to study the effects of encoding, but cannot be used to make conclusions on quality.

Dibrom has not submitted ABX results, but it's not hard to guess which samples he used to test (castanets, fatboy, at least). I've talked to him on this issue in IRC before (helping him to determine that the problem is caused by the encoding of 48kHz waves and not by the resampling process), and he said that the differences was too obvious to him that there was no need for an ABX test. That I will never doubt him because it's well known that he's a good listener and pre-echo is one of his specialties.

I'm sure if you ask him for an ABX (assuming he is free and available, which he currently isn't due to some personal problems), he will supply you with one.

There is no reason for Dibrom to falsify such results. Don't forget that this came at the time when Dibrom had a "hq" modifier in the presets, so that "--dm-preset hq standard" (when --dm-preset used to be what --alt-preset is today) does resampling to 48kHz in an attempt to REDUCE PRE-ECHO (the theory behind the reason why resampling to 48kHz should reduce pre-echo is very sound). It seemed ok on some tracks, but when Dibrom started testing it further, the problem with the 48kHz encoding starts to appear in more and more clips and finally he dropped the "hq" option.

Finally, I doubt there is very little question about Dibrom's integrity and capabilities. All you need to do is to look into the work and the results and the testings of the --alt-presets that this can't be the work of a conman who lies about the quality he hears.

[MaTTeR]

@tangent

I don't think anyone is doubting the integrity of his work at all. If they are then I missed it somewhere;) If it weren't for Dibrom's contributions we probably wouldn't have the nice --alt-presets

The fact is if I can't hear the artifact on a $1500 USD sound system then I'm happy with encoding @ 48khz. Like I said, I'm very picky with my audio quality but perhaps my ears are not as good as some others.

Anyone else hearing the pre-echo artifact when encoding @ 48khz? If so what bitrates are you using?

[tangent]

Apparently, the problems occur while Dibrom was testing his VBR preset, so it might be a problem which occurs only in VBR but not in ABR/CBR.

Well, I choose to resample to 44.1kHz, simply because I know that I will never distinguish between a cutoff of 22kHz and 24kHz (how many people can really hear above 19kHz? I already have trouble at 16.5kHz with pure sine wavs). I will probably never be able to hear bit errors from resampling. And it is much more likely, if not very likely, that in the near future, I will start hearing and recognizing preechos.

[DarkAvenger]

Don't forget that hifi equipment usually cuts off at 20kHz...

[tangent]

More from Dibrom from another forum:

Yeah... most of the information about the 48khz issues should be discussed within those threads. There was also some original discussion about this by 2BDecided on r3mix.net some time ago (I missed the original post) in which he found the exact same problem.

The issue seems to be that on some samples which normally exhibit pre-echo, and in which 48khz should reduce that by 10%, instead the pre-echo seems to be smeared across the stereo channels more.. it creates a different effect, often sounding worse.

In both my case and David's (2BDecided) we found that pre-echo was being smeared into the left channel and that overall the effect was more offensive than the pre-echo present at 44.1khz.

I tried multiple resampling methods including cool edit and ssrc, with and without dithering if I remember correctly.. it had no effect, so I'm quite certain the problem is not from resampling.

What I think is the case is that somewhere in the code, masking is being calculated improperly for 48khz. In fact, somewhere it may be hard coded to be optimal for 44.1khz which could mean that this effect might also occur at lower sampling rates as well. I haven't really tested this though due to a lack of time.

At any rate, the LAME developers were made aware of this problem some time ago, Robert and Naoki specifically.