Without hearing the 8 bit sounds it's hard to tell what kind of noise you are talking about, but there are a number of possible problems:
(1) there will always be a general loss of quality when converting from 16 bit to 8 bit - you're throwing away information, after all.
(2) if the 16 bit samples are not normalised (i.e. if they only use a small part of the full 16 bit range) then they will have very poor quality when converted to 8 bits - you should normalise first to make sure that you are using the full 8 bit range after conversion.
(3) there are two common but different 8 bit audio representations: signed (two's complement) and unsigned (offset), the latter being the more common. If you use the wrong format, i.e. converting to one format but then trying to play back assuming the other format, then the audio will be recognisable but sound pretty bad.
(4) for telephony (speech) applications it's common to use non-linear (logarithmic) encodings for 8 bit audio rather than linear - the two most common being A-law and µ-law. However your converted 8 bit audio is linear. If you're trying to use linear PCM in a telephony application that expects A-law or µ-law then it will just sound like a distorted mess.