Advances of these years in field of speech analysis, synthesis, and recognition are briefly reviewed.
First, engineering model of speech production is introduced as a time varying linear system drived by impulse train (voiced sound) or white noise (unvoiced sound). Some surveys are given on some important methods for speech analysis-synthesis such as linear predictive coding (LPC) and partial autocorrelation (PARCOR) as well as the most recent advances: line spectrum pairs (LSP) and composite sinusoid modeling (CSM). These are also fundamental techniques for narrow band speech transmission and efficient storage of speech.
Speech synthesis techniques including speech wave compilation, parameter compilation, and synthesis by rule are discussed.
Last, some varieties of speech recognition, that is, isolated word recognition, continuous speech recognition, and talker recognition are reviewed.
抄録全体を表示