To provide a method and a device for speech analysis that can perform real-time processing and are for unspecified persons.

A speech analyzing method is a computer executable and detects breaks of phonemes from an inputted speech. The speech analyzing method is characterized in: finding a plurality of frequency component data from a speech signal included in a plurality of specified-length time ranges by repeating a step of determining time in the inputted speech signal, a step of extracting a speech signal included in a specified-length time range from the time, and a step of decomposing the extracted speech signal into frequency component data; finding a plurality of correlativities from frequency component data corresponding to the speech signal included in mutually adjacent specified-length time ranges; finding a time range having a variation larger than two successive variations of variations of the plurality of correlativities; and sectioning the inputted speech signal into a plurality of parts based upon the time range.