RE-SPEECH RECOGNITION WITH EXTERNAL DATA SOURCES Russian patent published in 2019 - IPC G10L15/19 G10L15/02 G10L15/183 

FIELD: data processing.

SUBSTANCE: invention relates to means for producing transcription of a speech fragment. An initial version of the transcription of the speech fragment is obtained using an automated speech recognizer. Based on the language model not used by the automated speech recognizer when generating the original transcription, one or more terms which are phonetically similar to one or more terms which are already present in the original transcription version are identified. At the same time determination, whether phonetically terms are similar, provide definition of measure of similarity and comparison of measure with threshold, or determination, whether measure of similarity of measure of similarity relative to other pairs of terms exceeds. One or more additional transcription variants are generated based on the identified one or more terms. Transcription is selected from transcription variants.

EFFECT: technical result consists in improvement of accuracy of transcription of speech fragment.

20 cl, 3 dwg