An algorithm for joint suppression of noise and reverberation from speech signals is presented. The method requires a handclap recording that precedes speech activity. A running kurtosis technique is applied in order to extract an estimation of the late reflections of the room impulse response from the clap while a moving average filter is employed for the noise estimation. Moreover, the excitation signal derived from the Linear Prediction (LP) analysis of the noisy speech along with the estimated power spectrum of the late reflections are used to suppress late reverberation through spectral subtraction while a Wiener filter compensates for the ambient noise. A gain magnitude regularization step is also implemented to reduce overestimation errors. Objective and subjective results show that the proposed method achieves significant speech enhancement in all tested cases.
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.