I had an idea for a concept for restoring compressed audio and was curious about the thoughts others may have on its feasability and anything that anyone may may be aware of related thereto.

My idea is that the program would first scan your files to identiy specifics of an mp3 file, such as encoder used, bit rate, etc. and then scan a database (copyright issues may pose a probelm here) of some sort of data representation of the uncompressed data (using technologies like Music Brainz or musicmagic mixer's digital footprint method to get a correct match) and then use an algorythym; the application of which would be based the information from the database representing the uncompressed data and the user's particular method used to compress the mp3 to replace some portion of the audio data lost from compression.

Seems feasible to me, and its utility is obvious. Any thoughts?

first of all it's "algorithm" my apologies if you don't speak English that well. What you are talking about is already being done, what it is called is a "hybrid" coder. Wavpack does just this only it uses what's called a "correction" data along with the lossy file to reconstruct the original data so that nothing is lost. It's for those "anal retentive" audiophiles that can't make up their mind on whether to use lossy or lossless (it is very intuitive though and is less of hassle for some people). Database? maybe if you were talking about using a database to do feature extraction for classification of muscial instruments for machine learning, but I don't know

I speak and write English quite well. Thank you. Apparently, I have become too dependent on spell check. I realize that there is a way to create undo corrective data at the time of encoding an mp3, but my inquiry is rather directed at restoring audio data from a compressed file which was not originally encoded with something like Wavpac writing the corrective data embeded in compressed file.

EDIT: oh. You mean data compression.

writing correction data in a compressed file ?. Understand that comparing the two processes between lossy coding and lossless coding is like comparing an orange to an apple. Last time I checked when I had FLAC do residual dump for experimental purposes the data was quite large . (I can only imagine how wavpack works I will have to look into the technicalities of that).

[quote=HotshotGG,Aug 19 2005, 07:07 PM]
[quote]I speak and write English quite well. Thank you. Apparently, I have become too dependent on spell check. I realize that there is a way to create undo corrective data at the time of encoding an mp3, but my inquiry is rather directed at restoring audio data from a compressed file which was not originally encoded with something like Wavpac writing the corrective data embeded in compressed file. [/quote]

writing correction data in a compressed file ?. Understand the comparing the two processes between lossy coding and lossless coding is like comparing an orange to an apple. Last time I checked when I had FLAC do residual dump the data was quite large .

[/quoteI

I do realize this, and it is a fair point. As such, it would be most difficult to restore all the lost data, but it seems that it may be feasible to restore a portion of the lost data --the result of which would improve the sound of the file by not merely applying a dsp to create a perceived and subjective improvement but actually improve the sound of the compressed file objectively through the restoration of data lost through compression.

No I don't think that can be done . Psychoacoustics coding is much more complicated than source coding. What would you be restoring? quantization noise that you are doing psychoacoustics masking on?. The only thing that can be done, as you mentioned before is using dsp to do some form of noise shaping to reshape the quantization noise in time, that's all I can think of. I don't maybe an intgerized transform of some sort? The only application I could see that benefiting in is lossless coding. I have heard of the intMDCT (as in the bijective transform). Ohh wait I take that back for a concept I merily pulled out of my ass this might be possible.

http://www.idmt.fraunhofer.de/abteilungen_...er/icassp03.pdf

Just a thought. I know that a version of this concept has already been implemented in video, i.e. interpolating a scanned image or antialiasing fonts. Perhaps, something analogous could be conceived for compressed audio.

Sorry I edited my post above. Check out that paper it seems the guys at Fraunhoffer are hard at work on something like this. . To me though it looks like another hybrid though similiar to wavpack.

[quote=HotshotGG,Aug 19 2005, 07:48 PM]
Sorry I edited my post above. Check out that paper it seems the guys at Fraunhoffer are hard at work on something like this. . To me though it looks like another hybrid though similiar to wavpack.

[/quote

I am not sure I really understand it either, but it certainly bares relevancy. It does sound sort of like the wavpack concept.

Yeah it's really one or the other these days, I am sure they figured audiophiles are going to be anal retentive about the whole thing so that's one of the reason it's being developed. At this rate they are better off sticking with Wavpack as I can only imagine what liscensing fee's they are going to charge for this one (an arm and leg possible). I am not saying that out of spite either, because I understand they need to make a living and some of this stuff they are doing is very sophisitcated (what impact that involves on the end user though remains to be seen). It won't leave that much room open for researches and students though. Video is a little different ball game I don't know that much about it, just that it can tolerate more loss which I am guessing is the reason they used 2-pass quantizers and stuff.

"Audiophiles" will probably still claim to be able to hear the difference, no matter what improvements are made! What untold distortions due to this will they be able to "hear" through their SET amps I wonder?

Whoops, am I being cynical towards "audiophiles"?

If it would be possible, you would still have problems with the correct identification.
* Offset of the drive used to rip the music
* Slightly different mastering like volume adjustments for example on compilations

If Ogg Vorbis had usable bitrate peeling, an online music shop could use very high quality master encodings which it would strip down to the customer's desired quality. If the customer detects a flaw he could download only the missing data to reach the next higher quality level. Unfortunately this would be limited to specific music downloaded from that particular store.

Something that might be feasible is restoring lost (high) frequencies. Say your encoder cuts off at 16kHz, you could send a correction file with just the high frequencies, which should compress well (after fourier transform). But I sense great problems with offsets and sound near the crossover frequency.

All in all it would be much easier to get a new encode with higher quality, than to fix something which you have very little data about.

No, I consider myself a semi-audiophile at times. I think full-blown audiophiles are nutcases though so you are not being cynical at all. Stereophiles are even more off there rocker.




I very much agree with you on that part. You see the thing with Ogg Vorbis is that it's not writing an actual peeler that would be that much of a hassle. I am sure there are a few folks out there who have taken up task of doing so, including a company/developers who has posted to the board before called Vinjey Systems. The problem lies in the way the VQ data is organized in the backend. A lot of that has to be sorted out with the actual encoder. That's basically what the peeler is doing though it's making so many passes through the backend to dump any residue data that's not needed. The other problem is that I think I could be wrong, but in order to peel down to specific target level say from -q 5 down to -q 2 you would need an aribtrarily large codebook. I don't know if it could work in the opposite direction though the idea seems interesting. Edit: actually the idea has been brought up before

"Beni Cherniavsky mentioned a very intriguing counterpart to bitrate peeling. If you have a peeler that saves the bits it chopped off, you could reconstitute the higher quality files by adding the missing bits to the lower quality file. This idea could lead to a music download service where you can download a low quality preview version of a song, and if you are interested, download the missing bits to make it a high quality version." -vorbis.com