Investigation of a power distribution of a speech signal under empirical mode decomposition

A nonstationary signals analysis method which called the empirical mode decomposition is discussed in application to analysis of speech signals. Paper is devoted to the study of a question associated with power (energy) distribution of the speech signal in empirical modes which are extracted from this signal through decomposition. Material of researches is two speech signals corpora - English and Russian. Both corpora contain speech sentences that are pronounced by males and females. At the end of experiments we have taken the next results for two databases. Signal energy concentrates on the first four modes in English corpus, whereas one can observe the insignificant displacement and expansion of the energy concentration range in Russian corpus. We give the explanation of this effect from the point of view of statistical properties of the languages (speech phonemes). Investigation of the consecutive speech signal reconstruction has demonstrated that major portion of the signal energy (more than 90 %) is achieved under summing up all extracted modes except the two last ones. The last mode doesn't bring in any noticeable contribution to the energy of the recoverable signal.