基于卷积神经网络的实时语音分割优化研究
Research on Real time Speech Segmentation Optimization Based on Convolutional Neural Networks
杨波1
作者信息
- 1. 陇南师范高等专科学校数信学院,甘肃 陇南 742500
- 折叠
摘要
为进一步提高实时语音分割的性能,分析基于卷积神经网络(Convolutional Neural Networks,CNN)的实时语音分割优化方法.首先,介绍CNN的基本结构和在语音分割中的数学原理.其次,引入修剪技术,根据权重的重要性分数决定保留或删除权重.实验结果表明,该方法的准确率、召回率、F1值及用时均优于传统CNN.
Abstract
To further improve the performance of real-time speech segmentation, analyze the optimization method of real-time speech segmentation based on Convolutional Neural Network (CNN). Firstly, introduce the basic structure of CNN and its mathematical principles in speech segmentation. Secondly, introduce pruning techniques to determine whether to retain or delete weights based on their importance scores. The experimental results show that the accuracy, recall, F1 value, and time consumption of this method are all better than traditional CNN.
关键词
卷积神经网络(CNN)/语音分割/实时性/修剪技术Key words
Convolutional Neural Network (CNN)/speech segmentation/real time performance/pruning technique引用本文复制引用
出版年
2024
NETL
NSTL
万方数据