Modeling system and modeling method of hot-temperature-cold field data grading speech recognition model

Abstract

The invention discloses a hot-temperature-cold field data grading speech recognition model modeling system and a modeling method thereof. The system comprises an acquisition module used for acquiring a plurality of language samples; the classification module is used for dividing the plurality of language samples into a plurality of sample groups based on the retrieval probability of each language sample, and is connected to the acquisition module; the modeling module comprises a model storage unit for storing an initial voice recognition model and a training unit for inputting the language samples of the plurality of sample groups into the initial voice recognition model and training the initial voice recognition model to obtain a plurality of voice recognition models in one-to-one correspondence with the plurality of sample groups; the training unit is connected to the classification module and the model storage unit. According to the method, the problem that new hot data is easy to cut off due to the fact that all corpora are trained into one model by an existing speech recognition model is solved.

Description

technical field [0001] The invention relates to the technical field of speech recognition models, in particular to a modeling system and a modeling method for a data-graded speech recognition model in hot, warm, and cold fields. Background technique [0002] At present, the optimization of the speech recognition model is optimized for a specific field or scene. Generally, the text of a specific field or scene is collected, and then the speech recognition model of a specific field is trained to improve its recognition accuracy. Most of the training techniques of speech recognition models adopt the method of large-scale corpus and high-order N-gram. N-gram is a statistical-based language model in natural language processing. Its basic idea is to perform a sliding window operation with a size of N on the content of the text according to bytes, forming a sequence of byte fragments with a length of N. Each byte fragment is called a gram, and the frequency of occurrence of all g...

AI Technical Summary

Problems solved by technology

[0007] In order to overcome the defects existing in the existing technology, a kind of hot-warm-cold field data hierarchical speech recognition model modeling system and its modeling method are provided to solve the existing speech recognition model that trains all corpus into one model, and there is a new The problem that popular data is easily cut off

Images