Method for open creation of phonetic reading standard reference model

Abstract

The invention discloses a method for open creation of a phonetic reading standard reference model. A user becomes the expert user through program testing and system auditing to obtain the right and operations for the open creation of the phonetic reading standard reference model, and therefore multiple standard reference models are obtained for specific English text; when the user reads English, phonetic features of the user can be compared with the multiple standard reference models matched with the user to calculate scores; by normalizing and optimizing the scores, the equal-score errors caused when phonetic features of the user reading single letters and words are affected by context and the stress, tone, volume, pronunciation speed and the like are changed are reduced.

Description

technical field [0001] The invention relates to the technical field of phonological recognition and reading aloud scoring, in particular to an open method for creating a standard reference model for English reading aloud scoring. Background technique [0002] Today's academic experts agree that speech technology is one of the ten most important technological developments in the field of information technology in the near future. The application fields of speech recognition are very extensive. Scoring technical schemes such as likelihood scoring, logarithmic posterior probability scoring, segment classification scoring, segment duration scoring or its improved algorithm. [0003] The existing English reading recognition and scoring system is representatively developed by Carnegie Mellon University with Java language, a large vocabulary, continuous speech recognition system s Phin x 4;, by its front end ( Fron-tEnd), decoder (Deeoder) and knowledge base (Lin is i st) are join...

AI Technical Summary

Problems solved by technology

[0003] The existing English reading recognition and scoring system is representatively developed by Carnegie Mellon University with Java language, a large vocabulary, continuous speech recognition system s Phin x 4;, by its front end ( Fron-tEnd), decoder (Deeoder) and knowledge base (Lin is i st) are jointly implemented. The multiple standard reference models provided by sph in x 4 make the gap between program scoring and real expert scoring significantly narrowed, which is obvious The importance of the phonetic standard reference model for scoring English reading aloud, limiting the content of the English text to be read aloud, can still guarantee a certain accuracy rate, and the limited number of standard reference models is also difficult to adapt to a wide range of English texts, reading stress, pitch, volume and the diversity of pronunciation speed, English reading scores are often affected by the samples of the standard reference model of speech, resulting in uneven distribution of prior probability values ​​and large errors, because the quantity and quality of standard reference models limit English speech recognition to a certain extent The accuracy of scoring and the wider popularization of automatic scoring exercises and tests for English reading aloud

Images