Mongolian homophone acoustic modeling method based on end-to-end

The invention discloses a Mongolian homophone acoustic modeling method based on end-to-end. A Mongolian homophone model base, an acoustic model building module, a feature extraction module and a language processing module are used, wherein all Mongolian homophones are recorded in the Mongolian homophone model base; the Mongolian homophone model base comprises a localism model base and a standard model base; and the acoustic model building module also comprises a localism model base and a standard model base. The Mongolian homophone model base is built in advance; the localism model base and the standard model base are set in the Mongolian homophone model base; the standard condition probability of various homophones input by a user is calculated; the Mongolian is divided into two units including a standard unit and a localism unit, the feature comparison is performed according to the language types, and the standard condition probability is calculated, so that the recognition accuracycan be favorably improved