中大機構典藏-NCU Institutional Repository-提供博碩士論文、考古題、期刊論文、研究計畫等下載:Item 987654321/74653
English  |  正體中文  |  简体中文  |  全文笔数/总笔数 : 80990/80990 (100%)
造访人次 : 42691352      在线人数 : 1549
RC Version 7.0 © Powered By DSPACE, MIT. Enhanced by NTU Library IR team.
搜寻范围 查询小技巧:
  • 您可在西文检索词汇前后加上"双引号",以获取较精准的检索结果
  • 若欲以作者姓名搜寻,建议至进阶搜寻限定作者字段,可获得较完整数据
  • 进阶搜寻


    jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/74653


    题名: 高品質口述系統之設計與應用;The Design and Application of High Quality Spoken System
    作者: 朱祥豪;Chu, Hsiang-Hao
    贡献者: 資訊工程學系在職專班
    关键词: 口述
    日期: 2017-07-25
    上传时间: 2017-10-27 14:35:18 (UTC+8)
    出版者: 國立中央大學
    摘要: 本論文主要是研究基於神經網路之高品質口述系統的技術,並延伸相關的設計與應用,與以往最大不同的是,現今我們擁有更多的訓練資料、更快速的硬體設備、以及更多樣可搭配在語音合成的其他增強技術,讓合成語音的品質更加貼近真人聲音。要使這項技術能應用於生活上,需要設計具彈性且支援多方技術的工具來供實作,本系統主要是用Python語言開發,安裝在Linux作業系統上,需要一個可支援外部前端功能的工具,前端輸出的格式必須為狀態層次校準(state-level alignment)的HTS標籤,目前支援兩個語音編碼器(vocoder):STRAIGHT和WORLD,在訓練神經網絡之前,對語言特徵使用min-max正規化,而輸出聲學特徵則是採用mean-variance正規化。至於聲學建模(Acoustic Modelling)的原理,則是採用前饋神經網路(Feedforward Neural Network)和基於遞歸神經網路之長短期記憶(Long Short-Term Memory based RNN)於系統中實現。另外,就本系統的特色與長處,分別介紹三種相關的應用。最後,也期待這系統,除了不斷地在品質及效能上精進之外,也能推展到台灣各個有需要的地方。;This paper focuses on the technology of high quality dictation system based on neural network and extends the related design and application. The biggest difference is that we have more training materials, faster hardware and more Variety can be used in the voice synthesis of other enhanced technology, so that the quality of synthetic speech more close to the real voice. To make this technology can be applied to life, the need to design flexible and support multi-technology tools for implementation, the system is mainly developed in Python language, installed on the Linux operating system, you need a support for external front-end features Tools, the front-end output format must be state-level alignment of the HTS tag, currently supports two voice coder (vocoder): STRAIGHT and WORLD, before training the neural network, the language features using min-max regular And the output acoustic feature is normalized with mean-variance. As for the principle of Acoustic Modeling, the Feedforward Neural Network and Long Short-Term Memory based (RNN) are implemented in the system. In addition, the characteristics and strengths of the system, respectively, introduced three related applications. Finally, it is also looking forward to this system, in addition to constantly in the quality and efficiency on the sophisticated, but also to promote the various needs of Taiwan.
    显示于类别:[資訊工程學系碩士在職專班 ] 博碩士論文

    文件中的档案:

    档案 描述 大小格式浏览次数
    index.html0KbHTML339检视/开启


    在NCUIR中所有的数据项都受到原著作权保护.

    社群 sharing

    ::: Copyright National Central University. | 國立中央大學圖書館版權所有 | 收藏本站 | 設為首頁 | 最佳瀏覽畫面: 1024*768 | 建站日期:8-24-2009 :::
    DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU Library IR team Copyright ©   - 隱私權政策聲明