Artie Langston <artielangston@...>
I think the challenge is not actually in signal processing, but in some sort of artificial intelligence. Fldigi reads extremely well considering, when it's another machine doing the sending. With the peculiarities of various humans, on various keys, not so much.
This leaves the question, is designing an AI of that sophistication going to be worth the time and investment required?
I have noticed that my Digital Audio Workstation I use in my work as a musician can do amazing things with what is referred to as Quantizing a musical performance to various degrees. It can take a rhythmically imprecise musical performance and then "line it up" correctly with the beats and time signature, so it sounds pretty perfect.
Perhaps once the WPM is detected by the software, it could then quantize the CW before decoding, thus delivering something like W1AW or keyboard generated copy even from a shaky fist.