Speech recognition processing captures phonemes of words in a spoken speech string and retrieves text of words corresponding to particular combinations of phonemes from a phoneme dictionary. A text-to-speech synthesizer then can produce and substitute a synthesized pronunciation of that word in the speech string. If the speech recognition processing fails to recognize a particular combination of phonemes of a word, as spoken, as may occur when a word is spoken with an accent or when the speaker has a speech impediment, the speaker is prompted to clarify the word by entry, as text, from a keyboard or the like for storage in the phoneme dictionary such that a synthesized pronunciation of the word can be played out when the initially unrecognized spoken word is again encountered in a speech string to improve intelligibility, particularly for conference calls.