Vocal communication is most effective when the listener is able to observe the mouth of the speaker. This is especially true for the hearing impaired, and dramatically true for the deaf, who rely on lip-reading for comprehending speech. Communication over telephone lines is particularly onerous for the hearing impaired as visual information is unavailable. Our research addresses that problem by providing a computational means of taking speech as input and producing an animated mouth as output that moves precisely as if it were articulating the speech. In this paper we continue reporting on our progress in using moments of spectra-a measure of spectral shapes-to provide a direct mapping from the speech signal to parameters controlling the shape of the lips and position of the jaw during the articulation of the speech. The method requires no text nor does it rely on any form of speech recognition. We report in particular on the progress we have made in distinguishing the visemes-the visible phonemes-corresponding to /m/ and /n/.
展开▼
机译：当听众能够观察扬声器的嘴时，声音通信最有效。对于听力障碍而言，这尤其如此，对聋人来说，聋哑人来说，聋人们依赖于歌词来理解演讲。通过电话线路的沟通尤其繁重，因为视觉信息不可用。我们的研究通过提供作为输入和生产动画口作为输出的计算方式来解决这个问题，因为它恰好阐述了语音。在本文中，我们继续报告我们在使用光谱的矩 - 一种光谱形状的量度 - 以提供从语音信号到控制嘴唇形状的参数的直接映射和钳口在语音的铰接过程中的位置。该方法不需要文本，也不依赖于任何形式的语音识别。我们特别报告我们在区分探测器 - 可见音素 - 对应于/ m /和/ n /。
展开▼
