Inverse text standardization method and system

Abstract

The embodiment of the invention provides an inverse text standardization method. The method comprises the following steps: extracting keywords in a plurality of original speech recognition results concurrently output by a recognition module; performing scene division on the plurality of original speech recognition results based on the keywords, and determining scenes of the plurality of original speech recognition results; and respectively inputting the plurality of original speech recognition results into ITN resources corresponding to respective scenes, and outputting a plurality of inverse text standardized speech recognition results. The embodiment of the invention further provides an inverse text standardization system. According to the embodiment of the invention, under a divide-and-conquer calculation mode, the occupation of online ITN calculation resources is greatly reduced, and higher service capability is provided with lower cost. And standardized conversion of the special words of the user is supported more flexibly. For scenes which cannot be efficiently supported by certain scene rules, deep model optimization is carried out, and the problem which cannot be solved by the model can be avoided by adding the bad case into the rules of other scenes.

Description

technical field [0001] The invention relates to the field of intelligent speech, in particular to an inverse text standardization method and system. Background technique [0002] ITN (Inverse Text Normalization, Inverse Text Normalization) is a necessary processing process after intelligent speech recognition. Its function is to convert the original speech recognition results to a higher level, so as to be more in line with reading or reading habits or subsequent processing of other processes. Taking Chinese and English recognition as an example, the recognized texts are all Chinese and English words without numbers and symbols. But in terms of our writing habits, some expressions in standardized formats will conform to our reading habits, such as dates, percentages, phone numbers, etc. ITN is used to perform this level of standardized conversion on the recognition results. [0003] In the process of realizing the present invention, the inventors have found that there are ...

AI Technical Summary

Problems solved by technology

[0005] In order to at least solve the problem in the prior art that when the number of concurrency is high, the delay of ITN is high, which affects the number of concurrency supported by the maximum service

Images