06-19-2021, 02:17 PM
This can't be done without drastically increasing CPU usage. Currently we have a single convolution node for the entire sequence, which is very efficient. This suggestion would require convolution nodes for each instrument.