Michael Brebes wrote on Thu, 16 September 2010 18:29 |
Tomas Danko wrote on Thu, 16 September 2010 04:16 |
Michael Brebes wrote on Wed, 15 September 2010 22:08 |
J.J. Blair wrote on Mon, 13 September 2010 14:42 | Here's a question: All the examples on that page are 96 to 44.1. Are there less artifacts from 88.2 to 44.1? Is this a dangling integer thing?
|
From a purely mathematical point of view, there should be less artifacting with the 88.2 to 44.1 conversion since there is an even 2 to 1 relationship.
|
Again, this is a myth and not true.
This statement is founded on the misconception that it's merely a matter of stripping away every other sample, when in reality any SRC stage is a lot more complex than that no matter the original and target sample rates.
The math is the same regardless of 88.2->44.1 or 96->44.1 etc, simply put.
|
This statement is not founded on the misconception of stripping out every other sample. The advantage is that sample points are always going to be equally weighted against each other, instead of the positional balance of the sample points constantly changing, as in 96k to 44.1k conversion. That makes mathematically computation much more involved. I didn't take 3 years of calculus along with other higher mathematics for nothing. Please don't make broad statements unless you can back them up with more than "is a myth and not true", along with some poor assumptions.
|
We should hopefully get someone such as Bruno Putzey or Paul Frindle enter this thread and lay it down in a more precise and technical way, otherwise I fear the dispute will just keep going on.
Alexey Lukin said it, basically. And the way the best performing SRC happens nowadays practically speaking makes the issue you described above of no consequence. There are even other reasons for when a non-evenly weighed scenario is a benefit to the final outcome.
Also, finding a lot higher a sample rate for the intermediate stage gives several benefits related to implementation (i.e. filtering and such things, that makes for a lot greater impact to the integrity of the end result than anything evenly weighed per above).
Again, it's not just about less complex math but more to it than that. As always with digital audio, the Devil is in the details of implementation (and not picture perfect math theory).
Cheers,
Danko