On Wed, Jul 15, 2015 at 2:21 PM, James Bullock <[email protected]> wrote:

> On 15 July 2015 at 11:57:17, katja ([email protected]) wrote:
...
> For a library so systematic as LibXtract it makes sense to avoid
> wrapper code duplication. But will the extra complexity and overhead
> of Flext be justified in the case of these two wrapper externals plus
> corresponding makefiles?
>
> Do you mean runtime overhead or maintenance overhead? I’ve never set out to 
> measure the runtime overhead of Flext, but my impression is that it’s 
> insignificant and certainly I’ve never noticed it. But if there is evidence 
> suggesting otherwise, I’d be interested to know about it.
>
> In terms of maintenance / build complexity, since commit 73bd8015 Flext is 
> greatly simplified (through the use of C++ templates), requiring only a 
> single header include; no more building and link static libraries! This means 
> to support a flext external only two steps are needed: 1. add the Flext 
> repository as a git submodule, 2. include flext.h

Ow, if Flext complexity is hidden from the user nowadays I eat my
words (which were based on past experience).

> BTW, I’m planning to support a wider range of F0 detection methods, including 
> at some point your own Helmholtz algorithm.

Frankly I was thinking about how to split up routines in helmholtz~
such that it doesn't calculate a spectrum all for itself. That was
exactly the reason I looked at the design of LibXtract. But the point
with helmholtz~ is, I found that it needs to go back from
autocorrelation function to spectrum for refinement calculations in
the case of a high frequency fundamentals. The cascade approach would
not work here, I have to puzzle a bit more.

Katja

_______________________________________________
[email protected] mailing list
UNSUBSCRIBE and account-management -> 
http://lists.puredata.info/listinfo/pd-list

Reply via email to