Rule-synthesis of speech by word concatenation: a first step

Some preliminary work on rule-synthesis of speech by word concatenation is described. Words spoken in isolation were analyzed to extract their pitch, formant, duration, and amplitude parameters. The parametric representations of the words were concatenated and modified, and speech was synthesized from the modified parameters. An experiment was run to determine which parameters required modification to achieve natural prosody in the concatenated speech. The results indicated that the only parameter requiring substantial contextual modification was pitch; modifications of segment duration and formant coarticulation had minimal effects on prosody. These results were incorporated into a scheme for rule-synthesis of telephone numbers by digit concatenation. An evaluation experiment found the rule-concatenated numbers to be as comprehensible as naturally spoken numbers, and as pleasant as numbers resynthesized from naturally spoken numbers.