Least-squares inverse filtering of speech

Bradford Gover, James G. Ryan


The difficulty of dereverberating received speech in a hands-free system is examined. Dereverberation is difficult and imperfect even for cases where the impulse response from source to receiver is known. Furthermore, an inverse filter designed from the measurement of room impulse response cannot be used with a different source. If a very long filter is used, speech can be degraded far more than if no equalization has been attempted at all. Finally, the usefulness of least-squares inverse filtering for dereverberation appears to be limited.


Least squares approximations; Mathematical models; Signal filtering and prediction; Dereverberation; Least-squares inverse filtering

Full Text:



  • There are currently no refbacks.