That would be one very impressive DSP hack if you could eliminate the motor and other contact sounds. I highly doubt that you could cancel it out effectively but it is so tempting to try this to see how far you'd get.

A couple of approaches would be:

- bandpass filter in human voice range

- two mikes, only one facing outward, opposite polarity

- a lot of mechanical isolation

- a digital sample of the audio stored and subtracted from the input

And more such tricks. It would be a very interesting project to see how far you could push that.