8th International Conference on Spoken Language Processing

Jeju Island, Korea
October 4-8, 2004

Minimum Phase Compensation in Speech Coding using Hammerstein Model

Jari Juhani Turunen, Juha Tanttu, Frank Cameron

Tampere University of Technology, Pori, Finland

When using Bai's method for estimating the coefficients of both the linear and nonlinear parts of a Hammerstein model, there is no guarantee that the resulting linear filter will be stable. To obtain a stable linear filter the well-known minimum phase correction can be used. However, if it used without some compensation in the nonlinear part, one obtains a Hammerstein model with no theoretical foundation. In this paper we present a simple method for compensating the nonlinear part for corrections performed to the linear part. The compensation proved to be stable and fulfills the requirements of combined parameter estimation.

Full Paper

Bibliographic reference.  Turunen, Jari Juhani / Tanttu, Juha / Cameron, Frank (2004): "Minimum phase compensation in speech coding using hammerstein model", In INTERSPEECH-2004, 2685-2688.