To provide a method for generating voiced sounds from unvoiced sounds, without direct contact with the speakers and a voiced sound generation device which is useful to be used with the method.

A method for artificially generating voiced sounds from sounds containing speech information involves: a process to obtain the sound signals of sounds containing speech information and digitally convert the signals; a process to estimate voice tract characteristics from the signals; and a process to generate voiced sounds by filtering artificially generated periodic impulse signals using the estimated voice tract characteristics. The process of digitally converting the sound signals involves oversampling and the process of estimating the voice tract characteristics uses an adaptive digital filter.