In the context of telephony applications or movie sound recording, speech reverberation is problematic as it yields to a drop in both intelligibility and quality of the captured sound. Blind dereverberation is desirable, since the room impulse response is not known apriori or is changing unexpectedly. The use of higher order statistics (HOS) has proven effective in various areas of speech and signal processing, whenever dealing with a mixture of Gaussian and non-Gaussian processes or system non-linearity. The present paper extends on and offers a new variation of some of the existing work in the area of speech dereverberation using a Kurtosis maximization approach. The contribution of this work is in formulating the problem in the complex subband domain, based on new derived properties of the HOS, in a way that does not require an additional LPC analysis and residual filtering as with other methods. A complex normalized version of the kurtosis is used, as well as a variable adaptation factor based on the statistics of the error. The resulting scheme is effective in a number of tested scenarios and may potentially be extended to other audio signals.
https://www.aes.org/e-lib/browse.cfm?elib=18084
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.
Learn more about the AES E-Library
Start a discussion about this paper!
