Speech Synthesis of Classical Ryukyuan Language

琉球方言在日语中占有特殊的语言地位。由于研究琉球方言可以从其他方言所不涉及的特殊视角来观察日语，对日语的研究有很大的贡献，因此，我们利用数字信号处理技术，通过人工合成语音，对琉球语语音进行了准确的分析。通过开发能够合成各种琉球方言中的各种方言的通用语音合成系统，可以精确地分析各种方言。但是，这种通用语音合成系统在时间范围上存在局限性。该系统不能合成古典琉球语的语音。有用古典琉球语写的诗，如“大莫罗Soshi”，其中一些流传下来。为了在空间和时间上拓展对包括琉球方言在内的日语的研究，有必要研究古典琉球语。因此，在本研究中，我们开发了从古典琉球语文本合成语音的系统。利用这个系统，我们可以用模型构造法研究琉球方言。本文分析了《大牧歌》的部分传世诗歌的乐谱，并将其用马尔可夫模型表示出来，然后构建了一个自动生成《大牧歌》演唱声音的系统。实验结果表明，该系统能够正确地播放歌曲，准确率达到70%。因此，我们可以说，古典琉球语的法律是澄清与这种准确性。我们可以预期，非传承的歌曲也可以以这种准确度演唱。

Ryukyuan dialect occupies linguistically a special position in the Japanese language. Because we can observe the Japanese language from the special viewpoint, which is not involved in the other dialects, by studying the Ryukyuan dialect, we can contribute very much to the study on the Japanese language.Therefore, we have analyzed Ryukyuan speech accurately by synthesizing speech artificially using the technique of digital signal processing. And we have made it possible to analyze diverse dialects precisely by developing the general speech synthesis system which can synthesize each dialect of diverse Ryukyuan dialects.However, this general speech synthesis system has a limitation in time range. The system as it is cannot synthesize speech of classical Ryukyuan language. There are poems written in the classical Ryukyuan language such as "Omoro Soshi", and some of them are handing down. In order to expand not only in space but also in time the study of the Japanese language including Ryukyuan dialect, we need to study the classical Ryukyuan language.Therefore in this study, we developed the system in which speech is synthesized from a text written in the classical Ryukyuan language. By using this system, we can study the classical Ryukyuan dialect by the model constructive method. We analyzed the musical score of some handed-down poems of "Omoro Soshi" and expressed them by the Markov model, then, constructed a system which generates automatically singing voice from the test of "Omoro Soshi". As a result of experiment, the songs are replayed correctly by this system with 70% accuracy. Therefore, we can say that law of the classical Ryukyuan language is clarified with this accuracy. We may expect that non-handing-down songs can also be sung with this accuracy.