Subject: Re: [linux-audio-dev] plugin idea
From: Steve Harris (S.W.Harris_AT_ecs.soton.ac.uk)
Date: Mon Mar 05 2001 - 16:33:30 EET
On Mon, Mar 05, 2001 at 09:45:39AM +0000, Nick Bailey wrote:
> Are you /sure/ you want the drums sync'd to /everything/ that tightly?
> Wasn't it Mozart who encouraged a rubato independent of the left hand in
> his piano sonatas? Different genre, I guess. Anyway, almost certainly,
> a NN approach wouldn't quite work and you'd get the effect you desired
> anyway 8-)
Wow, um, I guess this guy is talking about phase effects between bits of
the same sound processed in different ways (ie. 1-100 samples sort of
thing) it can be anoying, and I've often thought that plugins should be
able to say how much latency thier running with (most could say if they
wanted to, I think). Then the host could fix this up if desired.
I dunno about neural nets though, last time I was working with neural nets
there were big problems when inputing temporal data (basicaly it wasa
giant hack). Not very appropriate.
You could solve this in LADSPA with a patten matcher:
______ _______
| | | |
---| P1 |--| |
|____| | Delay|
| fixer|
______ | |
| | | |
---| P2 |--| |---->
|____| | |
| |
______ | |
| | | |
---| P3 |--| |
|____| |______|
Just delay the input signals until you get a good envelope match across
the inputs, might be expensive unless you have a "learn" mode when you
just push an impulse (or a simple drum beat) through. I don't think any
current hosts can implement this network though.
It would be quite simple to implement something that did this with a learn
mode and bounded latencies. Fixing up more than a few ms is probably not a
good idea anyway.
- Steve
This archive was generated by hypermail 2b28 : Mon Mar 05 2001 - 17:00:34 EET