19 Oct
2016
19 Oct
'16
5:31 p.m.
On 10/19/2016 01:25 AM, jonetsu wrote:
On Wed, 19 Oct 2016 01:02:11 +0200 Robin Gareus
wrote:
For deep learning, the analysis tool itself would be the DAW in this
case which has context: All tracks, random data access, and it can
control the plugins.
There was an endeavor to standardize a very basic part: Allow a plugin
to analyze a complete track and write its own automation data:
http://lists.lv2plug.in/pipermail/devel-lv2plug.in/2016-February/001571.html
Something like this would be a very basic building block: the plugin
knows everything about itself, can do analysis and then "control" itself.
Even without fancy inter-plugin communication this could go a long way:
gates, adaptive de-esser, pitch-correction, noise cancellation, heck
even a channelstip plugin.
There are working prototype implementations (host & example plugin) but
so far it came to nothing. Go figure.
The actual use-cases were eventually implemented as VAMP-plugin
(Polarity optimizer) and plugin-internally (XT-TG Tom Gate).
ciao,
robin
Realistically it'll be a fixed set of plugin,
the analysis tool has
pre-shared knowledge about the available DSP + parameter behavior and
it is trained (neural network, heuristics, presets,..) specifically
for those plugins.
Yes, this is very likely how it will start to be known. At least it
will not be part of a specific DAW, but as a plugin, will be OK for
all DAWs out there.