Help - Search - Members - Calendar
Full Version: MP3 post processing
Hydrogenaudio Forums > Lossy Audio Compression > MP3 > MP3 - Tech
tacitus10
As many of you would know the Divx video codec, amongst others, has a post processing algorithm to help eliminate common artifacts associated with the compressor.

Why is it not possible to do the same for mp3? Audio processing can eliminate hiss and hum, why not mp3 arifacts?
userXYZ
How should the encoder know that there is an artifact or how to detect them? Video processing is a different thing which you can't really compare with audio compression.
QUOTE
has a post processing algorithm

Do you mean that post-processing option that should remove macroblock artifacts?

[edit] Even if you could erase artifacts with simple audio editing, you still would have to reencode the file, which would lead to new artifacts. [/edit]
NumLOCK
QUOTE (tacitus10 @ Feb 21 2003 - 09:22 AM)
As many of you would know the Divx video codec, amongst others, has a post processing algorithm to help eliminate common artifacts associated with the compressor.

Why is it not possible to do the same for mp3?  Audio processing can eliminate hiss and hum, why not mp3 arifacts?

When you compress a video, you don't expect it to look 100% the same. So you allow a huge compression ratio, and some macroblocks kick in. Then you can post-process the video with a heuristic algorithm, which does NOT improve quality, but it manages to hide many macroblocks, at the expense of a slight blurring of the image.

Audio is different; the encoder produces a datastream which had many differences in various frequencies, but should sound the same. If for some reason the resulting file does not sound the same (=artifact), there's no way for a decoder to know how it should sound. You can't just smooth sound like you smooth video. By the way, an audio artifact has nothing to do with a big square in video compression. Basically, most artifacts come when a sound just doesn't sound right, because your brain compare it with known instrumental sounds.

So, removing audio artifacts might be possible if, and only if:
- you know the exact nature of what an artifact is,
- or: the "artifact remover" has artificial intelligence, and a database of known instrumental sounds.

Cheers

[Edit]
The fact that video post-processing can be useful, also comes from the fact that there are MANY redundancies in a video image, even after compression with modern codecs. For this reason, post-processing can bring something. In audio, after compression it's very difficult to determine whether any specific modification will locally improve the sound (ie: closer to original), or make it worse. That's because of the way sound waves are.

This I think, derivates directly from information theory: when you post-process non-redundant (ie: optimally compressed) data, if you increase quality in some case, you'll decrease quality in another case.

On the opposite, when post-processing video data blocks, due to the redundancy of the images you will (maybe 95% of the time) be able to prevent a quality degradation.

If a perfect video codec would exist, no suitable post-processing would exist, let alone be useful for that codec.

The only useful post-processing for psychoacoustically compressed audio I'd think of, could be heuristic noise reduction. In some cases it could decrease any added "roughness" by the codec, but... in most cases the very nature of sound would get destroyed. So you see, it's not easy.
[/Edit]
tacitus10
Thanks for the reply. From being able to hear Mp3 artifacts myself, I though post processing might be possible. So I guess any 'smoothing' of the wave would only add a sine wave kind of artifact to replace mp3 artifacts.
niktheblak
Once audio data is gone (i.e. through overcompression, lowpass etc...), it's gone forever. Audio data can be "smoothed" out with lowpass, moving-point average filter etc, but that surely won't correct compression artifacts. Even hiss removal (through a template and Matched algorithm) is extremely difficult without removing useful data that "looks like" noise. Only very simple filtering like removing a 50/60 Hz ground-loop hum or excess high frequency noise are viable with audio data.

It's really unfair though. One can perform miracles with video through filters like Cnr2, Convolution3d, SmarthSmoother HiQ, WarpSharp and such. When using the right filters, recompression can actually improve video quality. I really, really wish such was possible with audio data...
NumLOCK
Yes, the biggest difficulty with audio data is to guess whether a given burst of (white) noise was part of the original signal, or not.

Also, audio is much more difficult to judge quality-wise.. artifacts are hidden. With video, every slight pixel difference can be seen (if the screen handles dark areas well), so you can more easily improve an image.

With both audio and video you can sometimes gain perceptual quality, but you won't gain information. If detail is killed, is killed.

However, you'd be surprised how many details can be recovered from a blurred image, for example. It can be astonishing !
Daijoubu
What about Sony Type-S DSP chip for improving ATRAC3 sound?
Med0
sure you can say that "DivX is blurring the image and that's how it hides macroblocks" but at the end, it looks better indeed!

one good example is Panasonic HE200 receiver (~350$USD) which has 4 DSP modes just for enhancing/resapling sound and the last one of four is for compressed sources. so, there ARE even consumer digital-to-analog converters and devices that have DSP chips for enhancing even compressed sounds.

but, I never heard of such software for PCs... sure there are many "enhancers" for WinAMP and so, but I'm not talking about that kind of stuff here.
Bedeox
Enhancing != Quality (for C impaired != means 'not equal')

You can try to 'enhance' sound using bass control. But will it gain quality? No.

Glitch removal could be used... something like deglitch for MP3, realtime, on the decoding stage.
It won't remove artifacts like preecho or too much lowpass.
It would improve only badly encoded MP3 with well-known 'ringing' artifacts.

(Anyway, I'll try tuning deglitch on some bad MP3s, just not now.)
ErikS
reviving a one month old thread... why not...

SBR which is used for mp3pro and aac+ is a postprocessing algorithm to enhance quality on the decoder side. It can fill in higher frequencies which were removed in the encoding, based on the lower frequencies and some extra clues provided by the encoder. So why would anybody say that postprocessing isn't useful for audio?
NeoRenegade
Your understanding of MP3pro is flawed. MP3pro isn't just MP3 played back on a decoder with postprocessing - it's a format encoded with most of the high frequencies removed, and then artificially reproduced on playback. Ultimately, the result can sound good, but it's definitely not hi-fidelity because it's destroying info and then replacing it with info that it "thinks" may be appropriate (which I'd bet isn't always the case as there should be some cases where SBR produces odd effects on the sound).
Med0
... and with this post you just proved how much do you know 'bout SBR smile.gif))
tangent
QUOTE (tacitus10 @ Feb 21 2003 - 04:22 PM)
As many of you would know the Divx video codec, amongst others, has a post processing algorithm to help eliminate common artifacts associated with the compressor.

Why is it not possible to do the same for mp3?  Audio processing can eliminate hiss and hum, why not mp3 arifacts?

The obvious artifacts found in video (and images) is quite well known (macroblocking, ringing), and there are algorithms available to do deblocking and deringing.

Audio isn't so simple. The only consistent artifact i can think of would be pre-echo, and detecting or removing that in the post-processing stage isn't trivial.
NeoRenegade
QUOTE (Med0 @ Mar 16 2003 - 12:58 PM)
... and with this post you just proved how much do you know 'bout SBR smile.gif))

Which is basically nothing.

I really loathe MP3pro. What the HECK were they thinking when Thomson decided to keep the .MP3 extension for MP3pro files?
Bedeox
Because it's somewhat compatible with MP3 and can be played by unmodified player?

SBR uses sound from file and duplicates whole decoded spectrum
at higher frequencies using additional data as a guide. Yes, it's fake.
Produces weird effects on music with lots of mid-freqs.
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.
Invision Power Board © 2001-2009 Invision Power Services, Inc.