Segment Inventory for Speech Synthesis

The inventory of segments required to synthesize an idiolect of American speech from recorded utterances has been partially investigated. A total of 43 phonetic units, including a unit of silence, is found essential. These units represent some of the major allophones of the phonemes of the idiolect. To obtain the segments, utterances are contructed with the desired sequence of the two phonetic units and with the proper intonation and stress environment. The utterances are recorded on magnetic tape, and the desired boundaries of the segments are determined by means of a technique employing a sound spectrograph. An attempt is made to match harmonics, formants, and amplitude envelopes between adjacent segments. To synthesize the idiolect it is estimated that an inventory of about 8500 segments is required. A sentence synthesized from over 40 such segments is demonstrated.