One of the results of the science of estimation theory has been the development of the linear prediction algorithms. This allows us to compute the coefficients of a time-varying filter which simulates the spectrum of a given sound at each point in time. This filter has found uses in many fields, not the least of which is speech analysis and synthesis as well as computer music. The use of the linear predictor in musical applications allows us to modify speech sounds in many ways, such as changing the pitch without altering the timing, changing timing without changing pitch, or blending the sounds of musical instruments and voices. This paper is concerned with the fine details of the many choices one must make in the implementation of a linear prediction system and how to make the sound as clean and crisp as possible.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.