| 摘要 |
PROBLEM TO BE SOLVED: To efficiently generate a sentence set for reading aloud sentences (text) needed to generate a speech database used for speech synthesis. SOLUTION: The total value of appearance rates of basic element pieces based upon tripartite phonemes as basic units is calculated as scores as to respective sentences in a sentence corpus (step 1-1). Basic element pieces which are already stored in the sentence set are excluded and repetitive basic element pieces in the same type in the same sentence are not included for score calculation. When there is one sentence with the largest scores (step 1-4), it is stored in the sentence set, and when there are a plurality of sentences with the largest scores, the total value of appearance rates of various extended element pieces based upon extended element piece units of syllables with environment, morphemes with environment, etc., is calculated by a similar method as to the plurality of sentences. Various extended element piece scores as to the same sentence are weightedly added, and regarded as composite scores, and the sentence having the largest value of the composite scores is stored into the sentence set. Task dependency can be increased by setting a weight coefficient of weighted addition. COPYRIGHT: (C)2004,JPO&NCIPI
|
