Method, system and platform for visual generation of speech recognition network

Abstract

The invention discloses a method for generating visualization of a speech recognition network, and the method includes: receiving keywords through a human-computer interaction interface. The current domain field is selected from a plurality of preset general domain fields, and each general domain field corresponds to a plurality of preset crawlers and corresponds to a plurality of preset web crawling pages. Get a general corpus. Get a specific corpus. Train the general corpus to obtain the general language model and the specific language model. After the WFST speech recognition network of the general language model and the WFST speech recognition network of the specific language model are connected in parallel, combined with the acoustic model and pronunciation dictionary, the WFST speech recognition network is synthesized through combination, determinization, and minimization operations. By configuring the system on the same platform, the training speed of the language model is accelerated, the product cycle is shortened, the labor consumption is shortened, and the labor cost is saved. At the same time, through the combination of general language model network and specific language model, the accuracy and efficiency of language recognition can be improved.

Description

technical field [0001] The invention belongs to the technical field of speech recognition, and in particular relates to a visual generation method, system and platform of a speech recognition network. Background technique [0002] At present, there are few relevant visual language model making systems on the market, and most language model making systems are customized at the command line level. The production of language models plays a pivotal role in speech recognition. Each speech company has its own team responsible for the model, but most of them are produced under the command line. In the prior art, the model customization process under the command line is uncontrollable, the version management is not easy, the risk is uncontrollable, and the process is not simplified enough. The reason for the above-mentioned defects is that the model is trained by manually inputting commands with various scripts under the command line. Manual training under the command line lacks c...

AI Technical Summary

Problems solved by technology

[0004] It can be seen from this that the visual speech recognition network used in speech recognition in the prior art has uncontrollable customization processes during the generation process, and the version is not easy to manage, and cannot satisfy multi-task language model training

At the same time, the visibility in model making is poor, and it is not convenient for multiple users to edit at the same time, thereby reducing the efficiency and accuracy of speech recognition model generation

Images