Andrew wrote:
> (hence, "wavelet"), and can reproduce the signal almost *exactly*
> by compositing the wavelets at playback. This breaks the bounds of
> Nyquist's rule, which states that you must sample at double the highest
> frequency you wish to represent... because you're no longer sampling.
>
I suspect this is nonsense. (sorry Andrew) You do have to sample the data to
get it into the digital domain, before you can process it. (Unless you are
going to do an ANALOG wavelet analysis - please supply diagrams)
> There are a few major drawbacks to wavelet compression... mainly the
> computational workhorsing needed, as well as the fact that the compression
> is unpredictable -- different waveforms will compress to different
> degrees,
> solely based on their structural composition. Also, it's not 100% real-
> time. You have to look at the signal over time to be able to give a
> wavelet
> representation.
>
I work with wavelet video compression, and that IS quasi real-time, ie you
have to operate on a whole video field at a time, so there is an effective
one field delay. I haven't got round to trying wavelets on audio yet, but
you would have to select some sample window size ( I think 20mS might be
appropriate ), so there will be a short delay. The computational demands are
probably similar to ATRAC. I suspect that wavelets will not be as good as
ATRAC for any particular data rate, as they seem to me less amenable to
psychoacoustic coding.
simon
-----------------------------------------------------------------
To stop getting this list send a message containing just the word
"unsubscribe" to [EMAIL PROTECTED]
