越南语文语转换技术探讨The study of Vietnamese Chinese language conversion technology 摘要:（摘要内容经过系统自动伪原创处理以避免复制，下载原文正常，内容请直接查看目录。） 跟着语音分解应用的普及，人们对分解语音的音质和天然度提出了更高的请求。越南语是越南的官方说话，属于南亚语系孟—高棉语族。越南语相似于汉语通俗话，也采取“声、韵、调”标志其音节。本文以开辟越南语文语转换(TTS)应用体系为目标，研究越南语语音分解，摸索进步音质和天然度的门路。依据越南语的特色，本文选用声母和韵母为分解基元，越语论文题目，并采取左、右相干的三音子模子。一个完全的TTS体系包含前端文天职析模块和后端语音分解模块。在前端文天职析方面，本文研究越南语的分词、词性标注、特别符号处置和韵律短语(L3)界限猜测等症结技巧。在后端语音分解方面，本文研究越南语练习语料的搜集、标注，和分解器的练习等。学位论文的重要任务包含：1。构建越南语的TTS基线体系，包含文本搜集和灌音、肯定音素列表、标注、高低文属性和成绩集设计、练习HMM分解器等。依据体系测试和评测的成果，从标注、高低文属性和成绩集三个方面临初步构建的分解器停止了改良。2。采取正向最年夜婚配法(FMM)和逆向最年夜婚配法(BMM)相联合的分词办法完成主动分词任务。提出了越南语词性标注计划，拟采取基于统计说话模子的办法完成词性主动标注。对越南语文本中涌现的特别符号停止尺度化处置，包含文本中的阿拉伯数字、数学符号、国际物理量和泉币单元、缩略语等。3。采取C4。5决议计划树算法完成越南语L3界限的主动猜测任务，包含属性信息提取、模子练习和L3界限主动标注，并对主动标注成果停止了主客不雅评测。试验成果注解：(1)采取本文计划改良后的分解体系，其分解语音的客观评测得分(MOS)分离为：集内3。89，集外3。55，解释改良后果显著。(2)运用论文中所述办法完成越南语主动分词和特别符号处置具有适用性和可行性。(3)采取本文所述办法猜测L3界限，越语论文范文，集表里测试的F一Score值分离为60。73%和57。96%，客观评测其可接收率分离为72。3%和69。0%。该办法用于越南语L3界限的猜测可知足语音分解体系的初步应用，且为进一步进步越南语语音分解的天然度奠基了基本。 Abstract: With the popularization of the application of speech decomposition, people put forward higher request to the voice quality and the nature of the decomposed speech. Vietnamese is Vietnam's official language, South Asian languages - mon Khmer languages. Vietnamese is similar to Chinese popular words, also take "sound, rhyme and tone" sign syllable. The to open Vietnam text to speech (TTS) system with the goal, research of Vietnamese speech decomposition, fumble progress quality and natural degree of roads. According to Vietnamese characteristics, the initials and finals for the decomposition of primitives, and take the three left and right tone coherent. A complete TTS system includes front-end and back-end module of text analytics speech decomposition module. In front-end Wentian job analysis, this paper studies the Vietnamese word segmentation, part of speech tagging, special symbols disposal and prosodic phrase (L3) limit speculation crux skills. The decomposition in the back-end voice, this paper studies the Vietnamese practice data is collected, annotation, and resolver practice etc.. The important tasks of the thesis include: 1. Construction of Vietnamese TTS baseline system and contains text collecting and recording, certainly a list of phonemes, mark, high and low, the properties and performance set design, practice HMM decomposition device. According to the results of the system test and evaluation, the paper has stopped the improvement from three aspects of the annotation, the high and low attribute and the result set. 2. Take the most positive mating method (FMM) and reverse the biggest match method (BMM) is combined with the segmentation method to complete the task of active participle. Put forward the plan of Vietnamese speech tagging, intends to take measures to complete the statistical model based speech POS tagging initiative. The Vietnamese language in the emergence of special symbols stop scale disposal, contains text in Arabic numerals, mathematical symbols, the physical quantity and currency unit, and abbreviations. 3. Take C4. 5 the decision tree algorithm complete Vietnamese L3 boundaries of active guessing tasks, contains the attribute information extraction, model practice and L3 boundaries of active annotation, and the active annotation results of subjective and objective evaluation. Test results notes: (1) to take this paper plans to improve the decomposition of the system, the objective evaluation score (MOS) is separated into: 3. 89, set 3. 55, explain the significant improvement. (2) the method applied in this thesis to complete the Vietnamese active participle and special symbols has the applicability and feasibility of disposal. (3) to adopt this method to guess the limits of L3, and the Score value of the F of the set table test is separated into 60. 73% and 57. 96%, the objective evaluation of its acceptance rate is 72. 3% and 69. 0%. The measures for the boundaries of Vietnamese L3 guesses contentment speech synthesis system in the initial application, and for the further progress of Vietnamese speech decomposition of natural degree has laid the foundation. 目录: |
