On detecting the steady state segments of phonemes by using the magnitude distribution of speech waveforms

To recognize continued speech, it is necessary to segment the connected acoustic signal into phonetic units. A new magnitude distribution is proposed as a parameter to detect steady-state segments in continued speech. The suggested parameter represents a change rate of the magnitude of speech signals. Comparing this distribution with the other in the adjacent frame, it is possible to determine whether the state of the frame is either steady or transient.<>