ASSESSMENT OF SPEECH SYNTHESIS TECHNOLOGY AS A HUMAN INTERFACE

1)合成语音质量的评估从1)可理解性，2)自然性和3)适合个别应用的角度进行。适用性可以根据应用程序目的、用户和环境来陈述。2)从可理解性研究中，我们发现：(1)双音节音素的可理解性分数比单音节音素的可理解性分数更重要，即双音节音素与句子的可理解性分数的关系比单音节音素的关系更密切。(2)单词的可理解性应该在美国哈斯金斯实验室的异常句子语境中进行测量。我们还发现了一种对异常句子进行感知判断的适当程序。3)我们发现，合成语音的自然度评价不仅要由普通人来做，也要由接受过语音训练的被试来做，因为自然度测试是衡量合成语音中实际语音表现的程度。4)根据合成语音在特定应用中的合适语音质量，对合成语音的适用性进行了实验研究。语音质量在平均频谱形状和基频方面发生变化。我们发现，合适的平均频谱形状取决于合成语音是在安静的房间里还是在办公室里被听到，适当的平均基频的变化取决于它是长时间使用还是短时间使用。

1) Assessment of synthetic speech quality has been made from the viewpoint of 1) intelligibility, 2) naturalness, and 3) suitability to individual applications. The suitability can be stated in terms of application purpose, the users, and the environments.2) From the intelligibility study, we found the followings: (1) Intelligibility scores of phonemes in bisyllables are more essential than the ones in monosyllables in the sense that the formers are more closely related to the intelligibility score of the sentence than the latters. (2) The word intelligibility should be measured in the anomalous sentence context as shown at Haskins Laboratories, U.S.A. We also found an adequate procedure for the perceptual judgments of anomalous sentences.3) We have found that the naturalness assessment of synthetic speech should be made not only by ordinary people but by the subjects who are trained in phonetics, since the naturalness test is to measure the degree of actual phonetic manifestation in synthetic speech.4) Suitability of the synthetic speech has been experimentally studied in terms of the appropriate voice quality of the synthetic speech for particular applications. The voice quality was varied in terms of the average spectral shape and the fundamental frequency. We found that the appropriate average spectral shape depends upon whether the synthetic speech is to be heard in a quiet room or in an office and that an adequate average fundamental frequency changes depending upon whether it is to be used for long hours or for a short period of time.