What you mentioned is indeed correct. I have seen people trying to work at around the telephone quality sampling rate - OK it sort of satisfies the guy Nyquist as the rate is just over twice the maximum frequency of 3Khz when you are talking about voice frequencies alone.

But that is not the end of all - if someone were to just sample at this rate without any additional signal conditioing at the input like the anti-aliasing filters, then you can well imagine the voice that you will reproduce! OK I would'nt want to speak into it.

I still maintain - if voice recording with 8051 is being done as a project to learn a few things - just go ahead. But if it is aimed at any serious end application or a sub stage in a whole product - you are making a mistake. And if you HAVE to do it then atleast use a mixed signal derivative from AD as already was suggested in a earlier post.

Raghu