Hey guys

I'm just wondering about the dithering features of Input and Output plugins in WinAmp. For instance, the MPC input plugin has the ability to dither the decoded output, and Peter's SSRC waveOut and DirectSound plugins can also apply dithering.

Using the output plugins, is dithering only applied when resampling the audio? All my encoded tracks have 44.1kHz, and I'm using the same sampling rate for output. Thus, I'm not really resampling, so is dithering not used then? Also, the output plugins allow for gaussian noiseshaping (i assume that is the highest quality of the three options?) ... when I am using this, with fast mode disabled, WinAmp is still taking 0% CPU useage, but jumps to 20-30% if I change the resampling rate on my Duron 800.

The MPC plugin also allows for dithering, but noiseshaping is disabled. Is this a feature that will be implemented in the future, or does it only apply to resampling?

I am in the process of moving over to MPC, but I still have a lot of MP3s in my collection from CDs that I ripped that I don't have access to anymore. I am curious what dithering/noiseshaping settings I should use to get the best results. If I can use dithering without a large CPU hit, I would consider using it even if I can't tell the difference. :angel2:

Yes.


Dithering is not used.


It might be implemented in future when decoding engine is updated.


If you have AC97 compatible sound card, you get best theoretical quality by setting input plugins to 24 bit with dithering enabled and set output plugin to resample everything to 48 kHz and 16 bits. Best dithering setting is triangular, I assume it's the same with noise distribution.
If you don't have AC97 sound card, just set best mode in input plugin your card supports and forget resampling output plugin.

I don't know what options it offers, but if there's anything like ATH Shaped, use that for noiseshaping.

--
GCP

First test versions had ATH modes, but those caused horrible distortion on some files and were removed.

Did you check out my comment about that? (The SSRC ATH dither needs a safety margin, the size of which is specified somewhere among all those weird numbers)

The problem manifested itself in my vorbis plugin as a sudden loud HF buzz.

--
GCP

If I remember correctly it was never examined, Peter just disabled the modes temporarily to be fixed later. But I do remember that the files it was tested on were loud, so probably this was the problem.

The problam was caused by samples reaching the maximum value (clipping). When dither was added, it resulted in an illegal value (more than maximum)

ok I'm still a little confused here... AC97 refers to the onboard sound chips on the motherboard, right?

I'm using a SoundBlaster Live Value, and i'm pretty sure it outputs to 44.1khz 16bit... although, resampling to 96khz 24bit still worked, but I didn't notice any difference (since the tracks themselves are 44.1/16)

maybe i should just forget dithering since it all sounds great already? ??? :ponder:

http://www.intel.com/ial/scalableplatforms/audio/
Audio Codec '97

blah blah blah


In short it is a specification. It is often used to refer to Generic AC '97 Compliant generic on-board sound chips.

I meant all AC97 compatible sound cards, they all resample.


Live! is AC97 compatible, if input signal differs from 48 kHz 16 bits, it will be resampled internally. Unfortunately, quality of resampling is far from ideal.

ATH shaped is only the best for a noise free recording, which is practically not possible.
For noisy recodings you must calculate the noise shaping floor from the ATH and the
noise floor of the recording. For very high noise floors the optimum is the shape of the
noise floor.

ahhh.... ok thanks!

so for best results, i should be using SSRC to resample to 48/16, triangular spectral shape, triangular distribution? (i think somebody said triangular was the best, even though there is an option for gaussian)

all AC97 cards do this? I thought it was a Live!-specific situation.

all AC97 cards resample, I think it's part of the AC97 spec.

hmm, interesting...now to find out if I should be taking this into account on my Aureal SuperQuad Vortex2 board...

Triangular distribution is the best, is better than gaussian. Triangular noiseshaping is is good, buth no noise shaping (~ "flat" noise shaping) is good enough, in my opinion.