Wow Arif!

What an interesting problem! This one will be fun to puzzle over.

I would start by autobauding each character, so to speak. Since these signals are likely to be manually produced there should be boatloads of time to process each input separately.

At 25 WPM (which I am assuming to mean Words Per Minute), assuming 5 letters per word, is 125 letters per minute. If one further assumes (off the top of my head, but could be looked up and/or calculated for better accuracy) 3 signal pulses per character one ends up with 375 data pulses per minute to process. That's 6.25 pulses per second, or a minimum of 160 mS per data pulse. A similar analysis would show that 5 WPM corresponds to a maximum of 1.25 data pulses per second, or 800 mS per data pulse. Thus, you have maximum and minimum periods over which to measure the period of any potential data pulses.

Thus, it should be fairly easy to measure the period of each data pulse. Compare the period of each pulse with the periods of the previous and subsequent n pulses, where n is an integer >= 1 (to be determined experimentally) and it should be fairly easy to bin them into long and short bins. This would allow you to reconstruct a sequence of dots and dashes.

Of course, this is assuming the person keying the data pulses is proficient enough to actually send Morse coded messages. And if they're not, then your machine will tell them that too, and pretty quickly I would think.

Once you've reconstructed the sequence of dots and dashes, you then need to parse them into whatever language the message was sent in. Since I've never learned Morse Code I'm not sure what a person does to make that translation, but in principle it should be doable. Or maybe this is one of those problems that is best done one step at a time. In other words, get the first part done, with a reconstructed sequence of dots and dashes, expressed as 1s and 0s, and then upload that sequence to a computer for further processing.

Joe