In a voice synthesis apparatus, by bounding a desired range of input text to be output by, e.g., a start tag "" and end tag , a feature of synthetic voice is continuously changed while gradually changing voice from a happy voice to an angry voice upon outputting synthetic voice.