Development of voice montage system.

这个主题的最终目的是实现一个声音蒙太奇系统，它试图用语音合成技术再现听者记住的一个人的话语。在这笔赠款的支持下，我们探索了实现这一目标所需的六个基本问题。我们为实现这一目标而努力的结果如下：(1)。与语音质量和说话风格相关的日常表达：通过对各种词汇来源的提炼，得到了1102种表达(2)。表达的声学相关性：对语音合成中尚未解决的声音年龄的声学相关性进行了广泛的研究，发现语音波形中的语速、音高轮廓和不规则性至关重要。(3)话语的记忆跨语言感知实验表明，对话语音质的记忆几乎独立于话语的语言。语音样本的呈现策略：发现视听交互很重要，说明语音与视觉的结合在实现语音蒙太奇中很重要(5)。合成语音的相似度评估：发现语音的音质相似度在记忆中保留了很长一段时间，但语音的说话人个性不一定在记忆中保持。灵活的语音合成技术：进一步改进了ARX分析合成方法，特别是针对GUI，让用户在语音合成方面有更灵活、更容易的操作。将上述结果整合到语音蒙太奇的实验系统中，该系统主要关注话语语音质量的再现。进一步阐述上面的问题(1)，特别是探索与说话风格相关的表达是实现项目最终目标所必需的。

The final aim of this theme is the realization of a voice montage system, which attempts to reproduce with the speech synthesis technique an utterance of a person that a listener remembers. With the support of this grant, we have explored six fundamental problems needed to achieve the aim. The outcomes of our endeavors toward attaining the goal are described below:(1).Everyday expressions associated with voice quality and speaking style :The elaboration on the extraction from various lexical sources resulted in 1,102 expressions.(2).Acoustic correlates of the expressions :Acoustic correlates of vocal age that have remained unsolved in speech synthesis have extensively been investigated and the speaking rate, pitch contour and irregularities in the speech waveform were found to be of vital importance.(3). Memory of the utteranceCross-linguistic perceptual experiments showed that the memory of the voice quality of an utterance was almost independent of the language of the utterance.(4).Presentation strategy of voice samples :Audio-visual interaction was found to be significant, suggesting an importance of the combination of the voice and visual display in realizing the voice montage.(5).Similarity assessment of synthetic speech :Similarity of voice quality of the utterances was found to be preserved for a long period of time but speaker individuality of the utterances was not necessarily maintained in the memory.(6).Flexible speech synthesis technology :The ARX analysis-synthesis method was further improved, in particular for GUI, allowing the user to have more flexible and easy manipulation in speech synthesis.The outcomes mentioned above were integrated into an experimental system of voice montage, which primarily focuses on the reproduction of the voice quality of the utterance. Further elaboration on the problem (1) above, especially on exploring the expressions related to speaking style is necessary to attain the final goal of the project.