Scoring method for re-scoring language model and speech recognition method
A language model, speech recognition technology
Active Publication Date: 2022-05-13
AISPEECH CO LTD
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0005] There are some methods to add OOV words that may be encountered by the second language model in the recognition process to the vocabulary, but this addition requires additional corpus, and the addition method requires retraining the language model, which is cumbersome
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0034] To make the object, technical solution and advantages of embodiments of the present invention more clear, the following will be combined with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are clearly and completely described, obviously, the embodiments described are part of the embodiments of the present invention, not all embodiments. Based on embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative work, are within the scope of protection of the present invention.
[0035] It should be noted that, in the absence of conflict, the embodiments in the present application and the features in the embodiments may be combined with each other.
[0036] The present invention may be described in the general context of computer-executable instructions executed by a computer, such as a program module. In general, program...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a scoring method for a re-scoring language model, comprising: pre-training a class-based language model as a re-scoring language model; determining the classification of OOV words to be scored; determining the OOV words to be scored according to a preset classification vocabulary and word frequency information parameter information; input the parameter information into the re-scoring language model; determine the probability of the OOV word to be scored according to the parameter information and the output of the re-scoring language model. The present invention pre-trains and obtains the re-scoring language model, and determines the probability of the OOV word to be scored by determining the parameter information of the OOV word to be scored according to the preset classification vocabulary and word frequency information and inputting it into the trained re-scoring language model. Re-scoring is achieved without using special UNK tags to replace OOV words, which completely solves the problem of word list mismatch and improves the accuracy of speech recognition.
Description
Technical field [0001] The present invention relates to the field of speech recognition technology, in particular to a scoring method and a speech recognition method of a heavy scoring language model. Background [0002] Automatic Speech Recognition (ASR) is a technology that converts a person's speech into text. Mainstream ASR systems generally contain a first-way language model and a second-way language model, in which the first language model is usually a statistical language model based on N-gram, while the second-way language model usually uses a neural network language model. [0003] The process of recognition is generally: the first language model first decodes the best N sentences, and then gives these sentences to the second language model for re-scoring. Because the second way language model will correct the score of the first way language model, so that the overall score is more accurate, so as to improve the recognition accuracy of the ASR system. [0004] In genera...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G10L15/01G10L15/02G10L15/16G10L15/183G10L15/22G10L15/26
CPCG10L15/01G10L15/02G10L15/16G10L15/183G10L15/22G10L15/26G10L2015/025
Inventor 俞凯戴凌锋刘奇
Owner AISPEECH CO LTD