| 摘要 |
<p>A method for speech analysis involves use of at least two band-pass signals for the speech signal, and for each of the part bands a discrete time prediction module for the response state is provided. On the basis of the speech signals, a common fundamental driver process is ascertained and a discrete time prediction module for the response state is selected so a driver-phase coupling function is obtained. The period lengths and further parameters of the driver phase coupling function are determined so a sequence of response states is generated by fundamental driver processes and approximate to the associated part-band. The period length of the driver-phase coupling function are used to classify the speech signal or part of the speech signal and in the case of voiced speech signals, the fundamental driver process and the driver-phase coupling function are used to ascertain a sequence of improved driver phases, in which the sequence of improved driver phases serves as a sequence of temporary driver phases. An independent claim is included for a device for carrying out the speech analysis method.</p> |
