Oh yeah, and-
HansP wrote on Thu, 28 April 2005 22:44 |
an AI algorithm could do some more about the distortion, it needs to model the dolby behavior to get down to a reliable transfer curve (histogram inversion etc).
|
Actually, I'm doing a version of that, Hans: at least I'm essentially making a histogram and keying off that for the expansion. The first try, I was remapping sample values to adjusted values (which can get very aggressive sounding), and the second time I didn't, because it tends to eat the reverberant field and peel sounds off it, moving them forward.
and-
Bill, a digital buss does not have its frequency response alter at different levels (that's part of the reason they don't sound as nice!) Neither does compression. Now, in each case there's a 'gotcha', as follows:
On a digital buss, if you're trying to reproduce seriously high frequencies, it'll take some headroom to do them properly. The reconstructed waveform can overshoot where the samples are. So, if you push digital past a certain point the highs start to shut down and to sound glarey.
With compression, if the release time is too fast it'll start to affect the bass. If you had compression with a super slow release time you'd get no alteration in frequency response, but as it speeds up it'll start to interact with the bass tonality. And since limiting is infinite-slope super-fast-release compression...
Apart from these reasons, gain changes in digital do not affect the frequency domain. (if I'm missing a phenomenon here, I'd be delighted to hear about it- I think I covered everything that can possibly happen)