MP3 often sounds bad as a format if you start with levels too high. Try saving your file so that the highest peak is no higher than -1dB.
The other problem of course is that you are encoding speech, and MP3 isn't exactly ideal for this, because of the way the encoding works. The masking system relies on relatively continuous bands of excitation where there is relatively little change between coding frames - and speech, with continuously changing levels and frequencies with gaps, doesn't conform to this pattern at all. So every artefact going is more likely to be audible than it would be with music. It's inevitable that this won't be so bad if you use higher coding rates - but the normal 128k is unlikely to be high enough to avoid them, I suspect.
I believe MP3 compression was designed for a target peak loudness of around -6dB for best results, while AAC does best with peaks at just under 0dB. Of course, compression and bit rate levels in the compression settings will certainly have an effect on the final output.
I believe MP3 compression was designed for a target peak loudness of around -6dB for best results...
Yes I think it was, but that was with music - speech has completely different peak loudness characteristics, and will generally encode more-or-less correctly at higher levels. When I was researching this as a Dip.Ac thesis relating to the accurate reproduction of noise levels, I discovered that almost anything other than music worked fine at a higher level...
And just to make it worse, some encoders do a better job of speech than others! Mentioning no lames - er, sorry, names...