A Speech Content Retrieval Model Based on Integrated Neural Network for Natural Language Description
Lifeng Wang · 2020
The current speech content retrieval model is confronted with the problems of slow retrieval speed and low efficiency. In order to achieve accurate retrieval of specified keywords, a speech content retrieval model based on integrated neural network for natural language description is proposed. First, we construct a sentence semantic knowledge base as a comparison criterion for retrieval. The speech and audio signals to be retrieved are input into the model in the same way as the keyword signals, and the signals are pre-processed through prefiltering noise reduction, signal enhancement and other procedures. Following the principles of deep learning algorithms, an integrated convolutional neural network is constructed, and feature vectors in the acoustic and language models are extracted and the similarity between the vectors is calculated. The retrieval result of the speech content is obtained by referring to the calculation result of the similarity. It is concluded through the test experiments of the model effect that compared with the traditional retrieval model, the retrieval accuracy of the designed speech content retrieval model is improved by 16.5%, and the retrieval speed has been improved to a certain extent.