I think you will find it very hard to get any sensible data out of an accelerometer without a lot of processing. It will also require running wires to every key, unless the intention is that the key actually hits the accelerometer at the end of its travel. It might be fine for a more percussive type of instrument. Determining if the key is actually up or down also gets a bit tricky.
The usual way to do it is to have a pcb with two contacts per key. A silicone membrane strip sits above it with two carbon pill contacts per key. One is a bit higher than the other, so as you press the key down, one switch closes before the other. You measure the time between them to determine the key velocity.
It's used by nearly all midi keyboards on the market, but I guess it's very hard to DIY that at home.
There are optical methods, like pnoscan.
http://www.midi9.com/products.htm based on reflective optical sensors like
http://www.digikey.com/product-detail/en/QRE1113GR/QRE1113GRCT-ND/965713Hall effect should work fine too, but then you need to put a little magnet in every key.