Development of Output System of Announcing Speech with Input of Kanji-Kana Sentences

组织该项目的目的是实现一个紧凑的规则合成系统，将输入的假名-汉字句子转换为自然发音的宣布语音。该系统由小型工作站和基于数字处理器的语音合成器组成。取得的主要成果如下：1.构建了一个词典，并将其存储在工作站的硬盘存储器中。词典的每个词项包括语素组成、音素表征、词性、变音/变化类型、重音类型和重要性指数。提出了一种语素分析方法，将输入句子分解成词串。该方法基于越长越好的词模板匹配策略，并对相邻词之间的连接词进行合适性检查。此外，还开发了一种语言分析方法，对由标点符号分割的表示…的单位进行句法和语义分析利用子句边界、直接成分与递归左分支结构边界等条件进一步得到句子的句法结构。在工作站上构建了一个程序，利用分词结果(包括重要性指数)和句法边界信息生成音素和韵律符号。研制了一种新型的终端模拟语音合成器，它由四个独立的级联通道组成，分别模拟四种不同语音类别和三种激励源，即声门波形、摩擦噪声和爆音噪声的声音传递函数。语音质量的显著发展主要表现在停辅音上。上述语音合成器的音节模板是为所有可能的日语音节构建的，并存储在工作站的硬盘存储器中。开发了一个程序，该程序将存储的音节模板连接起来，生成连续语音的特征参数的时间模式。采用目标函数的线性组合表示共振峰轨迹的方法，实现了元音和类元音的拼接。研制了一种模块化的小型终端模拟语音合成器，该合成器以声学特征的时间模式作为输入生成合成语音。该合成器由数字信号处理器和其他电子元件组成。将组件式合成器插入到工作站中，构成了一个总的播音输出系统。较少

This project was organized with the aim of realizing a compact rule-synthesis system which converts the input Kana-Kanji sentences into naturally sounding announcing speech. The system was composed of a small work station and a digital processor based speech synthesizer. The main results obtained are as follows :1. A word dictionary was constructed and stored in the hard disc memory of the work station. Each word item of the dictionary includes morpheme composition, phoneme representation, part of speech, type of declension/conjugation, accent type and index of importance.2. A method for morpheme analysis was developed which decomposes the input sentences into word strings. This method is based on the word template matching with longer the better policy, and checking of the propriety in the conjunction between neighboring words. A method for linguistic analysis was also developed which performs syntactic and semantic analyses for the unit segmented by the punctuation marks and represen … More t the obtained syntactic structure of the sentence using terms of clause boundary, ICRLB (immediate constituent with recursively left-branching structure) boundary, etc.3. A program was constructed on the work station which generates phone and prosodic symbols using results of word segmentation (including indices of importance) and the information on syntactic boundaries.4. A new type of terminal analog speech synthesizer was developed consisting of four separate cascade paths simulating the vocal transfer functions of the four different speech categories and three types of excitation sources, viz., glottal waveform, fricative noise and plosive noise. The outstanding development in the speech quality was observed typically for the stop consonants.5. Syllable templates for above mentioned speech synthesizer were constructed for all possible Japanese syllables and stored in the hard disc memory of the work station.6. A program was developed which concatenates stored syllable templates and produces temporal pattern of feature parameters of continuous speech. The concatenation of vowel and vowel-like sounds was conducted using the method based on the technique of representing formant trajectories by the linear combination of aim functions.7. A module-type small terminal analog speech synthesizer was developed which generates synthesized speech with the inputs of temporal patterns of acoustic features. The synthesizer was composed of a digital signal processor and other electronic components.8. The module-type synthesizer was inserted into the work station to construct a total system for the output of announcing speech. Less