基于卷积神经网络的语音识别方法设计与实现.docx

本科毕业设计 (论文) 题 目 基于卷积神经网络的语音识别方法 设 计 与 实 现 学 院 年 级 专 业 班级 学号 学生姓名 校内导师 职 称 校外导师 职 称 论文提交日期 I 基于卷积神经网络的语音识别方法设计与实现 摘 要 当前时代，科学技术蓬勃发展，计算机在各个应用领域都有所涉及，由于功能的便 捷变成了各行各业的“必需品”。而计算机最主要的功能就是信息的处理。根据信息自身 的类型和被接收对象来分，可分为图像和语音信息等类型，而如何将人能够理解的信息 交付给计算机处理就成了现代信息技术处理的关键。 语音识别技术，是解决计算机能够对语音信息辨别和理解的技术，作为人机交互技 术的一种，在发展历程中一直备受关注。与过往的语音识别技术相比，现代的语音识别 技术由于神经网络的出现而得以改进，神经网络也正在逐渐取代传统技术，将来在语音 识别技术领域，也必将成为技术主流。语音识别技术因其便捷性而被广泛应用，可以说 深入到各行各业之中，例如医疗服务，声控设备等等。语音识别技术在科技发展中，自 身也不断突破和提高。 在本系统中，关键部分是模型训练。使用中文语音数据集通过卷积神经网络进行声 学模型训练，使用该数据集对应的文本库进行语言模型的训练，结合这两个模型对语音 进行识别，通过解码器输出结果。 关键词： 语音信息 语音识别 神经网络 模型训练 Ⅱ Design and Implementation of Speech Recognition Method Based on Convolution Neural Network Abstract In nowadays, modern science and technology have vigorous development, computers are involved in different application areas, because of its convenience of functions, it has become a necessity for all walks of life.And the main function of the computer is information processing. Depending on the information itself and the type of object received, it can be divided into types such as image and audio information, and the method of transmitting human-readable information to computer processing is the key to modern information technology processing. Speech recognition technology, which solves the problem that computer can distinguish and understand speech information, has been paid much attention in the development process as a human-computer interaction technology. Modern speech recognition technology is improved by the advent of neural networks compared to conventional speech recognition technology, and neural networks are gradually replacing the traditional technology, and speech recognition technology will become mainstream in the future. Speech recognition technology is widely used due to its convenience.It can be said that it goes deep into all