[vorbis-dev] audio analysis tool?

[alexander lerch]

Hi!

A value that gives you how much information is in a file is
the entropy of a signal. But this value is interesting for
lossless coding and not very meaningful with lossy coding
algorithms.
Johnston introduced (in 1988 IIRC) the term perceptual
entropy, which does more what you want, because it gives you a
value/bitrate at which the audio material can be coded without
perceptual loss. However, for the calculation you have to use
a perceptual model and thus it will not be a better estimation
than what audio codecs do in the encoding process. But of
course you could output the data of the psychoacoustical
analysis of any encoder and calculate the perceptual entropy
(which is also quite similar to what audio codecs do,
especially at VBR-modes). BUT, I doubt that the perceptual
entropy of a (high quality) coded signal will differ much from
the perceptual entropy of the original file (if there is no
big bandwidth and stereo image loss).
To determine if a signal was coded or not is -without a
reference wave file- very difficult. It is even more difficult
if you cannot make assumptions about the encoding algorithm. I
would consider the spectrogram view as good point to start.

Cheers,
Alexander

Eric Seppanen wrote:
> On Wed, Nov 06, 2002 at 02:13:39AM +0100, Segher Boessenkool wrote:
> 
>>Eric Seppanen wrote:
>>
>>>I'm curious, does anyone know of a way to analyze a wave file to determine
>>>the quantity of information in it?  For example, given a particular .wav
>>>file, how can I find out if it's an original or was encoded to something
>>>lossy and back?
>>
>>Just look at a spectrogram.  Big empty regions at the high frequencies
>>==> lossy.  Lots of other features, too (it's not very hard to see which
>>codec you're looking at).
> 
> 
> That's helpful, but I guess I was imagining that there might be some way 
> of quantifying how much data is there, for instance having a tool that 
> could look at an audio input and notice that it never seems to have more 
> than roughly 128kbps of information in it.
> 
> Since vorbis has quality settings, I figured it might be able to quantify 
> how much signal (not signal magnitude, but quantity of frequency data) is 
> coming in.
> 
> The other thing that occurs to me is that some codecs may "fake" 
> high-frequency data when decoding just to fool the uneducated listener.  
> While a sharp person could probably spot this in a spectrogram, I wonder 
> if it would be possible for a tool to detect "faked" data and discard it 
> when measuring the quantity of "useful" data.
> 
> I'm wondering how difficult it would be to create such a tool using the 
> vorbis encoder as a starting point.
> --- >8 ----
> List archives:  http://www.xiph.org/archives/
> Ogg project homepage: http://www.xiph.org/ogg/
> To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
> containing only the word 'unsubscribe' in the body.  No subject is needed.
> Unsubscribe messages sent to the list will be ignored/filtered.
> 


-- 
dipl. ing.
alexander lerch

zplane.development
http://www.zplane.de
holsteinische str. 39-42
D-12161 berlin
 fon: +49.30.854 09 15.0
 fax: +49.30.854 09 15.5

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.