A CELP-Bsased Speech-Model Process

摘要

A new Speech-Model Process (SMP) is proposed as a test signal for speech processing applications to avoid several shortcomings (e.g., speaker dependencies) of using natural speech as a test signal. The generation procedure for this SMP, which is based on a modified CELP-structure, is presented. The controlling part of this SMP involves several Markov chains to adapt the time-varying properties of the process to those of natural speech. Furthermore, we show that this signal can also be used as a suitable test signal for telephone speech-applications, if we limit it to the telephone frequency-band.