Method and device for matching voice recognition text with command word text

Abstract

The invention provides a method and a device for matching the voice recognition text with the command word text. The method comprises the steps of acquiring text through voice recognition; performingword segmentation on the text to generate a plurality of text words; performing word segmentation on the preset command word text to generate command words; determining a command word with the highestsimilarity with each text word, and then generating a first corresponding relation; based on a highest similarity, determining the word similarity between the text and the command word text; determining the pinyin of each word in the text and the command word text; determining the pinyin of each words in the text with the highest degree of similarity relative the pinyin of each word in the command word text, and then a second corresponding relation; based on each highest pinyin similarity, determining a highest pinyin similarity between the text and the command word text; based on the word similarity and the pinyin similarity, determining the similarity between the text and the command word text. According to the invention, the annotation of data is not required. The method and the deviceare high in tolerance degree to voice recognition errors. Meanwhile, the follow-up matching process of command words is guaranteed.

Description

technical field [0001] The invention relates to the field of recognition, in particular to a method and equipment for matching speech recognition text and command word text. Background technique [0002] In the current voice interaction, first use voice recognition technology to recognize the voice as text, and then perform corresponding processing. In this process, it is often necessary to match the voice recognition text with the functional command words to determine the user's goal; However, in practical applications, due to problems such as too short functional command words, unclear meanings, and inconsistent with conventional expressions, the recognized text will be quite different from the actual one. [0003] At present, in general text matching algorithms, the similarity between texts is generally considered directly. There are some current text matching algorithms: [0004] A, string level, such as based on (weighted) edit distance, retrieval model, etc.; [0005...

AI Technical Summary

Problems solved by technology

In the case of a certain error in speech recognition, basically all existing methods will have problems; among them, as far as the string level is concerned, sufficient semantic information cannot be matched, and in the case of a certain error rate in speech recognition, the failure rate will be very high; and for the shallow semantic level, constructing shallow semantic analysis such as dependency syntax analysis will consume more processing time, and this will lead to a reduction in the efficiency of the entire online environment, which is not very practical ; and based on shallow semantic analysis, there are also certain differences in the adaptability to colloquial (even speech recognition errors) expressions, and the same is true for building similar dictionaries and then comparing them at the string level It cannot solve the problem of speech recognition errors; as for the deep semantic level, a large amount of labeled data is required, and in such a relatively new field, the acquisition of labeled data is extremely large, or even directly impossible in a short period of time of

[0009] Therefore, in the process of interacting with the robot, the user needs to speak the function command word to be able to match and enter the corresponding function. However, based on the above method, if the voice recognition information is insufficient, the voice recognition result will be There is a large deviation

Images