Larry Smith has presented the most complete and reasonable discussion
I've heard of analog audio to MIDI conversion. The genetic approach
did not exist when I began to survey the possibilities, and it appears
to offer the necessary loops. I am just now reading through the last
year's correspondence and will probably run across many letters on the
subject, but I would like to note the concepts I have identified with
their plusses and minuses.

1. Artistic recreation: A capable musician listens to the audio and
recreates the performance on a MIDI recording. Musical ability to
interpret the original recording and correlate from memory to the MIDI
file and editing time are limitations.

2. Sonogram output edited into note events: Extremely messy data due
to overlapping harmonics and loud notes blocking soft ones make this
totally impractical in manual translation. [I have examples of
sonograms of Ampico rolls that don't correlate even though they're the
same data!]

3. Signature analysis through separately recording note signatures (or
estimating them from solo audio notes) then performing stepwise
identification of individual notes through process iteration: Notes do
not play the same twice in harmonic content and decay. The big and
separate notes come out marginally well, but the remainder are hidden
in a field of hash that does not identify.

4. Number 3 but with subtraction of identified notes before analysis
for the next note: This still leaves a field of hash after picking out
the big ones.

5. Any or all of the above played back against the original audio by
binaural comparison to assess the accuracy by time and stereo location
shifts: This helps in the correlation, but gives no clue as to what
technique is needed to make a fix.

The genetic analysis technique effectively does item 3 and 4 if I
understand it correctly, and can unravel the Gordian knot with enough
machine time.

The next question is, what hardware and software are necessary to give
this a trial run? I have lots of overnight time available on DOS and
UNIX workstations at home, and have my own resident AIX administrator
and C-writer, but he has not osmosed the genetic algorithm recommended
reading list.

Ok, Larry, Tag, you're It =).

Karl A. Petersen¶
3116 N. Hearth Ave.¶
Meridian, ID 83642¶
208) 884-3488 h¶
(208) 323-7124 w