Solutions for a very high number of inputs

I actually designed something similar to this several decades ago.  First, some general comments:

1.  Don't worry about "near real-time response".  MIDI only runs at 31.25 kb/s, so you're going to have some delays unless you use time-stamping.  A small amount of delay is actually good because real musicians don't play notes at exactly the same time and if you synthesizer does it will sound artificial.  The same goes for pitch -- I've heard synthesized music sounds more natural if instruments are slightly out of tune with respect to each other.  Of course, you don't want to go too far: "What's the difference between the first chair viola and the last chair viola?  Answer: A measure and a semi-tone."

2.  Do worry about switch bounce.  Mechanical switches do not produce clean on/off transistions.  They tend to bounce between on and off until the contacts settle.  You'll need to filter out these transistions.

So let me tell you what I did back then.  Since the technology was rather different, I'm going to borrow from The Four Yorkshiremen.  I wanted to build a simple polyphonic electronic organ using two 37-key surplus keyboards, with a one-octave overlap to give the usual 61-key organ keyboard range.  These keyboards didn't have switches: "Your keyboard has actual switches?  Pffaaa -- I had to make my own switches out of paper clips with wire insulation stuck over them, and glue each one to the end of a key -- I even had to make my own glue."  (OK, I made up the part about making my own glue.)

Each key had a wire-wrap wire soldered to it and the wires were connected to the inputs of five 74150 16-to-1 multiplexers.  "Pffaaa -- back then we didn't have CMOS or even LS.  In fact, Schottky hadn't even been born yet."  Basically, the five 74150's plus a 74LS151 acted as an 80-to-1 multiplexer to select a key.  An advantage of TTL is that you don't need pull-up resistors on inputs: they automatically float high if not connected.

Now the interesting part: I needed to debounce the inputs but I didn't want to add 80 resistors and capacitors.  Besides, debouncing with RC networks requires that the digital inputs have hystersis (Schmitt triggers) which 74150s don't have.  So I did the filtering digitally.  I used a 256x4 bit SRAM as a bank of 4-bit counters and then time-multiplexed logic to update the counters so that each counter was visited every 250 us or so.  Each visit would read a key's counter value into an up/down counter, increment or decrement the value depending whether the key was pressed or not (saturating at 0000 or 1111), and then write the updated value back into RAM.  If the MSb changed, it meant that the key had been pressed or released long enough for the bouncing to settle.  When MSb changed, the board would interrupt my computer (a Heathkit H-8) which would update my synthesizer board to add or delete a note.  The keyboard scanner was a couple dozen TTL and LSTTL chips and ran at 2 MHz -- a nice solution for 1977 wire-wrap technology.

A scheme similar to this might work for you.  This being the 21st Century, I'd get a microcontroller board with lots of I/O pins such as an ST Discovery board and then add external multiplexers to get your 300 inputs down to 16 or 32, whichever works for your board.  You can do the filtering for 16 or 32 inputs in parallel.  If you don't know the trick, let me know.

John...those were the days when ingenuity was really tested, bought back memories...Rod

John...those were the days when ingenuity was really tested, bought back memories...Rod