Vocal expression of emotion : Comparison of Japanese and English

在2002年4月1日至2005年3月31日期间，我们对日语和美语发声情感表达的声学、发音和知觉特征进行了不同的实验，结果以期刊或后续出版物的形式发表(见所附参考文献)。主要结果如下：(1)对美国英语和日语发音和声学自发言语的语料(由松木NTT实验室录制)的检验表明，悲伤和模仿的悲伤话语(A)不同于阅读话语，(B)具有相似的声学特征(高F0，改变F1和改变音质)，(C)但发音不同(例如，嘴唇、下巴和舌头)。此外，被听话者高度评价为悲伤的言语与高F0和改变的音质有关。(2)对不同说话风格的句子表达话语的发音和声学特征分析表明，节奏模式和音调轮廓都不同，重音模式(由音节和边界大小表示)通常独立于语调工作，例如，为了引起怀疑，突出音节(/Won/)实际上具有最低的F0值。(3)对日本和美国听众进行的关于日语副语言信息的知觉研究表明，母语为母语的人能够在三维知觉空间中正确地识别副语言信息的类型，但非母语听众表现出不同的模式，这取决于他们对语言的熟练程度；此外，回归分析显示声学测量与知觉空间高度相关。

During the period from April 1, 2002 to March 31, 2005, various experiments were done on the topic of the acoustic, articulatory and perceptual characteristics of vocal expression of emotion in Japanese and American English, the results have been published as journal or proceeding publications (see attached bibliography).The main results are the following :(1) Examination of articulatory and acoustic spontaneous speech American English and Japanese speech data (recordings made at NTT Laboratories, Atsugi) indicate that sad and imitated sad utterances (a) are different from read utterances, (b) have similar acoustic characteristics (high F0, changed F1 as well as changed voice quality), (c) but are different in terms of articulation (e.g., lip, jaw and tongue). In addition, speech rated highly by listeners as sad is associated with high F0 and changed voice quality.(2) Analysis of articulatory and acoustic characteristics of expressive utterances of a given sentence in different speaking styles suggested that both rhythmic patterns and voice pitch contours vary, and that the stress pattern (represented by syllable and boundary magnitudes) often works independently from intonation, e.g., for suspicion, the prominent syllable (/won/) actually has the lowest F0 value.(3) Perception studies with Japanese and American listeners of Japanese paralinguistic information indicated that native listeners could identify types of paralinguistic information correctly in 3-dimensional perceptual space, but non-native listeners showed a different pattern, depending on their degree of proficiency in the language ; moreover, regression analysis showed high correlation between the acoustic measures and perceptual space.