Continuously adaptive linear predictive coding of speech

摘要: A continuously adaptive approach to speech encoding is presented. In contrast with other adaptation methods, it provides reliable modeling of the transition between two phonemes, and, unlike usual block-stationary techniques, eliminates need detect these transitions. Continuously linear predictive coding takes into account inherent nonstationarity signal by using expected minimum rate change model parameters as a constraint in recursive estimation parameters. The criterion considered constrained-least-squares cost functional which incorporates equal weight all instantaneous errors up time observation. An appropriate algorithm given, and simulations are presented illustrate basic cost-performance tradeoffs involved approach. >