Domain language model construction method and device, computer equipment and storage medium

Abstract

The invention discloses a domain language model construction method and apparatus, a computer device and a storage medium and belongs to the technical field of speech recognition. The method comprisessteps of converting a universal language model into an equivalent first WFSA network; screening out an optimal path meeting a preset condition from the first WFSA network according to a preset numberof domain corpora so as to construct a second WFSA network; and normalizing the second WFSA network, and converting the normalized second WFSA network into a domain language model. The method is advantaged in that under the condition of insufficient domain training corpora, a domain language model which meets a specific scene and has universal generalization ability can be quickly constructed.

Description

technical field [0001] The invention relates to the technical field of speech recognition, in particular to a method, device, computer equipment and storage medium for constructing a domain language model. Background technique [0002] Speech recognition schemes are mostly recognition schemes based on language models. When training a language model, the most commonly used model is the N-Gram model. The N-Gram model is a statistical language model. Generally speaking, the larger the corpus, the better the effect of the model. As the scene continues to deepen, it is often required to make various language models that meet the needs of specific scenes and have generalization capabilities, which puts forward higher requirements for the selection of corpus. [0003] At present, there are generally two commonly used methods for constructing language models that meet specific scenarios. One is to directly collect relevant field corpus for training, and the other is to integrate th...

AI Technical Summary

Problems solved by technology

[0003] At present, there are generally two commonly used methods for constructing language models that meet specific scenarios. One is to directly collect relevant field corpus for training, and the other is to integrate the trained language model with a general language model according to a certain weight. To increase the generalization ability, and the above two methods require a large amount of domain training corpus, but it is not easy to find domain corpus that fits the scene

Images