Keyword searching method based on MVGG-CTC

Abstract

The invention discloses a keyword search method based on MVGG-CTC. The method comprises the following steps of acquiring voice data; the voice data being preprocessed; inputting the voice data into an MVGG-CTC model for training; constructing a voice model and a dictionary; converting a pinyin sequence output by the MVGG-CTC model into continuous characters; constructing a voice keyword search library by using the inverted index; inputting keywords in the voice keyword search library for retrieval; and obtaining a retrieval result output by the voice keyword search library. The keyword search method based on MVGG-CTC is advantaged in that (1) the network structure is improved, and efficiency and accuracy of speech recognition are improved; (2) robustness of the model is improved; and (3) intelligent automatic identification of rapid keyword retrieval can be realized.

Description

technical field [0001] The invention belongs to the technical field of speech recognition, and in particular relates to a keyword search method based on MVGG-CTC. Background technique [0002] With the rapid development of network communication technology, audio and video media has become a mainstream form of information dissemination, and its characteristics of efficient circulation and fast interaction have become more and more prominent. The dissemination and interaction of public opinion information has reached an unprecedented level. While bringing convenience to people, the negative impact of the dissemination of bad public opinion information has become more and more significant. The dissemination of public opinion information such as pornography, terror, and violence not only Violation of public order and good customs also brings huge threats and hidden dangers to public security, which is also a major concern of relevant departments in our country. How to effective...

AI Technical Summary

Problems solved by technology

[0005] 2) Traditional acoustic models include traditional GMM-HMM (mixed Gaussian model-hidden Markov model), HMM-DNN (hidden Markov model-deep neural network model), etc., and these models are cascaded by multiple models Composition, not only reduces the efficiency but also reduces the accuracy in the cascading process

[0006] 3) Although the WFST (weighted finite state decoder) in the prior art has good speed and accuracy, it still needs to train each module separately when it is applied to the model, because the model is complex and some key information will be lost in the intermediate link. The results are often unsatisfactory

[0007] 4) The most common retrieval technology is the relational database. Most of the model software has applied the relational database. Its update query is relatively good, but it is not good at processing large amounts of data.

Images