Abstract
The visually impaired encounters problems when reading a document through a speech synthesizer. One of these problems is that it takes long time to read a document from top to tail, and the other problem is that he can not understand the position where he read. This paper describes an improved speech representation techniques by using both one-dimensional sound source and skipping-over function. An experiment shows that the proposed method could reduce total speech output time by skipping over unimportant texts, and that soundsource-based position prediction techniques was effective for users.
