METHOD AND SYSTEM FOR GENERATING TEXT REPRESENTATION OF USER'S SPEECH FRAGMENT Russian patent published in 2020 - IPC G10L15/22 

FIELD: data processing.

SUBSTANCE: group of inventions relates to natural language processing. Method includes: receiving an indication of a user's speech fragment; generating at least two hypotheses; generation of a set of paired hypotheses by an electronic device from said at least two hypotheses, one of which contains a first hypothesis paired with a second hypothesis; determining for one pair of paired hypotheses from a pair of paired estimates of pair estimation; generating a set of features of a speech fragment indicating one or more characteristics associated with that fragment of user speech; ranking the first hypothesis and the second hypothesis based on at least a pair estimate and a set of features of said speech fragment and selecting a first hypothesis as a text representation of that fragment of the user's oral speech if the first hypothesis represents a hypothesis with the highest rank.

EFFECT: technical result is generation of a text representation of a fragment of user's oral speech taking into account user characteristics and acoustic properties of a speech fragment.

20 cl, 7 dwg