SIMULATION OF VOICE QUALITIES IN SPEECH

语音质量模拟

• 批准号: 2733685
• 负责人: INGO R TITZE
• 金额: $ 22.97万
• 依托单位: DENVER CENTER FOR THE PERFORMING ARTS
• 依托单位国家: 美国
• 财政年份: 1995
• 资助国家: 美国
• 起止时间: 1995-07-01 至 2000-06-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/2733685
• 关键词: artificial intelligence; behavioral /social science research tag; biomechanics; computed axial tomography; computer program /software; computer simulation; electrical measurement; human subject; magnetic resonance imaging; psychoacoustics; respiratory airflow measurement; respiratory imaging /visualization; speech; speech synthesizers; vocal cords; vocalization; voice

This research addresses one of the remaining challenges in speech science, high quality speech simulation. We define speech simulation as a form of speech synthesis in which the movement of air and tissue is under experimental control, rather than the resulting acoustic signal. From the early days of speech synthesis nearly a half century ago, the expectation has always been that a better representation of the laws of physics of air and tissue in motion would produce better synthesis. Although this expectation still exists today, the payoff has been slow, primarily because there are few data sets from which to build theoretical generalizations. In this proposal, the principal investigator and his colleagues draw upon experience gained with simulation of the phonatory processes to include the entire vocal tract in sentence-level speech production. The first phase will be to obtain naturalness in speech quality that is comparable to formant synthesis by modeling a few specific speakers from whom extensive data sets will be available. The second phase will be to develop scaling and modification rules that will allow the voice of a given speaker to be transformed into a different age, gender, emotion, and voice quality. The transformation will also include induced or corrected voice and speech disorders. The idea of voice transformation (conversion) is not new, but the attempt to do it all in the articulatory domain is relatively untried. The results will have practical and theoretical impact on the development of assistive devices for voice/speech impaired populations, for surgery performed on the larynx and upper respiratory tract, and for speech training and rehabilitation.

这项研究解决了语音科学中其余的挑战之一， 高质量的语音模拟。 我们将语音模拟定义为 语音综合空气和组织的运动在下面 实验对照，而不是由此产生的声信号。 来自 言语合成的初期将近一个半个世纪前，期望 一直以来更好地表示空气物理定律 运动中的组织会产生更好的合成。虽然这个 今天仍然存在期望，收益很慢，主要是 因为很少有数据集可以构建理论 概括。在此提案中，首席调查员及其 同事通过模拟语音获得的经验借鉴了 在句子级的语音中包括整个声带的过程 生产。 第一阶段是获得语音的自然性 通过建模一些特定的特定的质量与共振剂合成相当 将提供大量数据集的演讲者。 第二个 阶段将是制定规模和修改规则，以 给定扬声器的声音要转变为不同的年龄， 性别，情感和声音质量。转换也将包括 引起或纠正的声音和语音障碍。 声音的想法 转型（转换）并不是什么新鲜事物，而是尝试全部尝试的尝试 枢纽域是相对未经试验的。 结果将有 对辅助设备开发的实用和理论影响 对于语音/语音受损的人群，用于在喉部进行手术 和上呼吸道，以及言语训练和康复。