I'd record my own, but obviously it would be great if there was a default option.

This is totally doable.
perhaps with partial failover to text to speech for unassigned samples?
My 2p

Ah ha, I have just stumbled across baby smash and this was the first feature I was going to request this very feature.. I would definitely take the time to record the sounds.. a simple directory 'speech' or something where you could record and deposit a named mp3 such as 'A.mp3' and when the A key was pressed, if the A.mp3 was present that would play, if not then text to speech is the default

Ohhh, Could be cool ...
Support for multiple languages ... then I could/will translate it to danish :-)
mvh

If you make the recording "wizard" super straightforward, this could go fairly quickly.
A wizard-like: Show the letter on the screen, then prompt the user to hit a key to begin the recording, same key to end the recording, automatically play back the recording to verify then same key to go on to next letter. Different key to re-record, and the other options to stop the recording process. Any un-recorded items would just use the default text-to-speech.

Giving the chance to have the letters read out in the native language would be nice.

I think most people would have the patience to record all the sounds - it's a one time only job afterall.

Be great if Parent could record all the letters - thereby providing the correct translation.