mandel wrote on Tue, 18 October 2005 09:38 |
To sum (correct me if I am wrong)...
if we record and store at 24/48, the data playback should theoretically be indistinguishable from 24/192 or 24/384, using Nyquist and Fourier laws...more info does not mean more accuracy or resolution here, since Fourier analysis says all one needs is frequencies up to the Nyquist frequency to break down all audible signals into integral of sine curves...The only difficulty, and the reason why digital can sound bad, is in the implementation of filters and other errors...if "perfect" filters are used and quantization errors are neglible, then sound reproduction should be "perfect" - no different from that in a "live" stage...
am I right here?
|
Well the system is only going to be as good as the mics and speakers, but you get the general idea.
mandel wrote on Tue, 18 October 2005 09:38 |
so, the question...I'm not challenging Nyquist and Fourier, but Fourier analysis is after all mathematical theorem, not a physics one...so could it be possible that using wavelet technology as complement, sound reproduction could be improved...better filters or easier filters to implement, and/or since wavelet does not deal with sampling rate, could there be something outside sampling rate that one could measure and could better represent the ear-brain thingy?
or would you argue that perfect is already perfect?
|
Filters and Fourier work very well together, so I doubt wavelets would make life any easier there.
It is possible that a model which more closely matched the ear/brain response, in both its strengths and weaknesses, would allow a better capture system, however I don't find it very likely because
1) The system would inevitably be more complex than a constant rate sampling system, complexity tends to create more problems.
2) We can already sample what most research tells us is more than the neccessary bandwidth with noise and distortion levels lower than the other links in the chain presently achieve, and perhaps will ever achieve.
So barring some radical shakeup of our knowledge of audio perception, I would say that constant rate sampling will probably remain the best option as an interface between analogue and digital.
IMHO A wavelet based model is more likely to be of use in lossy compression, signal analysiz and processing, and sound synthesis. So basically things you do to the PCM stream in between sampling it and playing it back.