I don't know about Gossett, but Dex and I did this on our Cook-Morrill

trumpet interface, and I think the CNMAT/Gibson/Zeta project did this

as well.

On ours, we fired a noisy attack on one MIDI channel as soon as we 

detected playing over a threshold.  Then we waited a fixed time (like 

100 ms) and collected pitch info from many sources (valve info, lip

tension, and acoustic features from two mics).  Then we fired a best-

guess pitched noteon on a different channel, and updated that with 

bend messages.  We sent a note off on the attack channel.  There

was no perceived latency, but there was occasional monkey business

on the pitch update side.

Prc

Ref:  

On Mar 24, 2013, at 9:00 AM, chuck-users-request@lists.cs.princeton.edu wrote: