基于影像和语音分析的发音器官运动可视化

项目名称： 基于影像和语音分析的发音器官运动可视化

项目编号： No.61273288

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 杨明浩

作者单位： 中国科学院自动化研究所

项目金额： 81万元

中文摘要： 目前，语言病理学家和言语治疗师们对发音障碍进行诊断和康复评估时，通常依赖个人经验做出判断，其主要原因有两点：一是医学界对人类的发音机理尚不完全清楚；二是多数发音器官隐藏在人体内部，器官运动与语音的关系难以观测，缺乏客观评估依据。本课题以同步分析器官运动和对应语音之间的关系为出发点和目标，建立形状约束的发音器官三维运动离散紧凑表示和三维精细控制模型，研究器官二维残缺影像轮廓的三维形状复原方法，从低质量医学影像中重建发音器官的三维运动，实现发音器官的协同控制和可视化。在此基础上，对比典型患者与正常者语音及器官三维运动的差异，在时序上深入分析发音器官运动缺陷与异常语音的关联，在合作医院的支持下，将可视化工具和病理语音映射模型在临床中验证，提高科研成果的实用性。本课题的开展，除了为语音障碍诊断和康复训练提供客观评估依据外，也为进一步研究发音障碍的生理机理提供直观的可视化分析工具和重要的数据支持。

中文关键词： 多模态病理语音数据库；医学图像处理；语音分析；语音可视化；嗓音评估

英文摘要： In spite of great progress have been achieved in voice medicine recently, the situation has not changed that the diagnosis for language disorder patients depends on the personal experience of speech pathologists and speech therapists. There are mainly two reasons behind this phenomenon. (1) How speech is produced is not completely clear in physiological mechanism for speech pathologists; (2) Most speech organs are hided deeply in body, which makes them difficult to observe, so that objective voice analysis in dysphonic patients is still a huge challenge for speech pathologists and speech therapists. This study analysis the relationship between organs' movements and voice features firstly and then elaborately construct the 3D model of speech organs from the low-resolution medical video and drive the vocal tract movements based on input voice. Based on the visualization technique for speech organs movements, this study further construct the mapping model between organ movements and pathological voice by analysising and comparing the difference of organ movements and voice between language disorder sufferers and normal persons. And a speech organs' movements and voice feature database will be constructed depending on the mapping model. Finally, supported by cooperative hospital, we will verify the visualization too

英文关键词： Multi-modal pathological voice database；Medical video analysis；Speech analysis；Speech visualization；Voice measurement