Development of a speech act identification system considering prosodic, voice quaility and linguistic information

我们构建了一个包含语言、韵律和副语言信息的自然会话对话语音数据库。提出了一种新的呼吸语音质量声学参数。从数据库中提取单音节感叹词（如“un”、“oh”、“haa”），分析说话风格与言语行为之间的关系。一组表达不同含义的感叹词在网上可用。对语音行为的自动识别也进行了评估，表明所提出的声学特征的有效性。对重复的感叹词（如“unun”）也进行了分析，表明重复次数与副语言功能之间的关系。对问句型话语进行了分析，揭示了短语末语调与人际关系之间的关系。

We constructed a natural conversational dialogue speech database, including linguistic, prosodic and paralinguistic information. A new acoustic parameter was proposed for breathy voice quality. Monosyllabic interjection utterances (such as "un", "oh", "haa") were extracted from the database, and the relations between speaking style and speech acts were analyzed. A set of interjections expressing different meanings were made available in the web. Automatic identification of speech acts were also evaluated, indicating the effectiveness of the proposed acoustic features. Repeated interjection utterances (such as "unun") were also analyzed, indicating relation between the number of repetitions and the paralinguistic functions. Question-type utterances were also analyzed, indicating relations between the phrase final intonation and the inter-personal relationship.