New Lossy Audio Codec, JPEG is not an ideal audio codec Options

[rutra80]

3. Save as 16bit 44khz wav files
4. Convert to lossy
5. Convert back to wav (using foobar)

I wonder what effects would give different sampling-rates and lowpass values, and turning on/off fb2k's dithering...

WV - I see no diffrence from oryginal picture (best audio picture codec :] )
OFS - the are any diffrences too !!!!!!!!!!!!!!!!!

Because they're lossless codecs, I wonder how their hybrid modes would perform though...

This post has been edited by rutra80: Sep 16 2005, 20:19

[de Mon]

MPC - increase contrast and saturation of images
MP3 - on the contrary decrease contrast a bit
OGG - do perfectly from all above and show oryginal colors
WV - I see no diffrence from oryginal picture (best audio picture codec :] )
OFS - the are any diffrences too !!!!!!!!!!!!!!!!!

I wonder how these results correlate to real double blind listening tests.
Can anybody ABX samples? May be we have got a new lossy audio quality measuring method

[XoR]

Because they're lossless codecs, I wonder how their hybrid modes would perform though...

I tested lossy mode
To me WM and OFS sound better even than same bitrate OGG so that's not surprise to me that they "look" better

This post has been edited by XoR: Sep 16 2005, 21:46

[HbG]

I think its quite a surprise the relative order of encoders seems to hold on images.

[Shade[ST]]

I think its quite a surprise the relative order of encoders seems to hold on images.

A greater surprise would be to see the compression gain factor they actually reach on such samples. These bitmaps are coded as 22050Hz sampling rate / 8 bit per sample, correct? what compression ratio is reached, while maintaining discernable image quality? Such an analysis would be interesting, as well as seeing what type of image compresses the most (gradients, checkered pattern, etc..)

[Mac]

You can get some nice visual effects by applying your own distortion instead of encoding to mp3.. flanging and multi-tap delay are quite nice

[JB_]

While studying for a Fourier Analysis test some of my flatmates and I were discussing how well JPEG would encode music. Since both lossy audio codecs (MP3, Vorbis, etc) and JPEG operate on the same basic idea (discarding unimportant data in the frequency domain) we decided it would be an interesting thing to test.

That simply won't work well - because audio coders exploit the irrelevancy according to the human psychoacoustics, adding noise in frequency regions that are masked by outer-inner ear transfer and inner-ear processing.

Good audiovisual coders exploit the visual irrelevancy - so, you will end up with noise allocated in regions that do not correspond to psychovisual masking critereia.

yes, but the objective is the same in either compressors. the goal is to look as close to the original picture or to sound as good as the original sound.

With this technique only the methods are switched.

By using a picture compressor in a sound file, you're trying to find which is the compressor that makes the file looks as good as the original, depending on the settings you use. By using a sound compressor in a picture file, you're trying to find which is the compressor that make the file sound as good as the original (which i think hasn't been done yet)

When settings are specified to the compressors to work (or compress) the file is reconstructed based on those same settings. how come an audio compressor knows that an given file is an audio file? the irrelevancy model are part of a tentative of recreating the file. All the compressors do is try to reproduce the original file. To make an analogy, its like modern painters trying to reproduce a famously known piece of art, which reproduction stays closer to the original one.

The compressors aren't reasonful, they don't make decisions, they use algorithms to work, regardless of its content/data. With this I think theres no reason not to think that an jpeg compressor can't be looked as an audio compressor. In the end it only shows how good it is at reproducing the original file.

[pepoluan]

Holy thread resurrection!

You know that the thread is 2005, don't you?

But... I do wonder how the result would be with today's hunked up lossy compressors...

And no one tested AAC yet...

[Febs]

Maybe we can post the results on Head-Fi and then people there can use the quality of pictures to decide which encoder to use for audio.

[...Just Elliott]

THIS IS ART.

[Woodinville]

Holy thread resurrection!

You know that the thread is 2005, don't you?

But... I do wonder how the result would be with today's hunked up lossy compressors...

And no one tested AAC yet...

Somebody ought to test AAC for image coding.

After all, look for Kuo's paper on using SNS (Spatial Noise Shaping) in image coding

[Gabriel]

Somebody ought to test AAC for image coding.

After all, look for Kuo's paper on using SNS (Spatial Noise Shaping) in image coding

Sure, and PNS would reproduce film grain

[gaekwad2]

Too bad Nero discounted the use of PNS without testing its usefulness for images.

Anyway, here are some results from the current NeroAAC encoder:


(crops taken from a 3mp image (original), size refers to the whole picture (original is 1322KB))

[Woodinville]

Too bad Nero discounted the use of PNS without testing its usefulness for images.

Anyway, here are some results from the current NeroAAC encoder:


(crops taken from a 3mp image (original), size refers to the whole picture (original is 1322KB))

:rollingonthefloorlaughing:

There was an irony, I'm told, to the review of Kuo's paper. One of the reviewers mistook the use of a full-image FFT (for SNS purposes) as an argument to use a full image FFT for coding. Talk about missing the point completely...

[gameplaya15143]

Wow, I have got to try this at home

[gaekwad2]

Vorbis -q-2 looks nice as well:



At higher bitrates AAC, Vorbis and mp3 look fairly similar.

[Mr VacBob]

Can't let this topic die, so I wrote yet another one!
http://astrange.ithinksw.net/tools/imgpcm.c

This is fancier, not that it helps anything; it converts RGB to YUV, with Y in the center and U-V (the hue, sort of) as the stereo difference, then lays the image out along a Hilbert curve.

Lossless:

(The colors are already wrong since it pretty much stores the image by dropping the saturation)

lame 32kbit:

(the mp3 is 235kb)

And, uh, aotuv b5:


Apparently it can't handle the signal phase changing.

No idea how to save the image saturation without something gross, like an NTSC signal...

[cabbagerat]

Can't let this topic die, so I wrote yet another one!
http://astrange.ithinksw.net/tools/imgpcm.c

This is fancier, not that it helps anything; it converts RGB to YUV, with Y in the center and U-V (the hue, sort of) as the stereo difference, then lays the image out along a Hilbert curve.

That's a really cool tool, thanks.

I can't believe this thread is alive after nearly four years

[gaekwad2]

Can't let this topic die, so I wrote yet another one!

[...]

No idea how to save the image saturation without something gross, like an NTSC signal...

How about multichannel?

Instead of Lame you'd have to use fhg mp3surround, but apart from that it should work.