Improvement of Speech Clarity in Noisy and Reverberant Conditions Based on Active Noise Control and Inverse Filter Algorithm

Abstract

In an indoor electronic meeting system, active noise control (ANC) system is generally used to reduce the ambient noise and improve the clarity of the received speech signal. Typical ANC methods like the filtered-x least mean square (FxLMS) algorithm attempt to control the background noise, which is a kind of additive noise. However, when the speaker of ANC system is used as the speech communication speaker, the nonadditive noise like reverberation, dispersion and attenuation caused by the propagation path is also necessary to be considered. As a common phenomenon that interrupts the communication, the speech reverberation is focused on in this thesis. To obtain the better approximation of the speech reverberation in the given room, the reverberation is simulated by image source model method (ISM) and added in the propagation model. Then, the ANC-IF (active noise control-inverse filter) method that combined the FxLMS algorithm and the FIF (fast inverse filter) algorithm is proposed to control the noise included the background noise and the speech reverberation in the electronic meeting system. The performance of the FxLMS algorithm in the reverberant condition was first investigated. The noise source produced the factory noise as the background noise, and the ANC speaker generated the opposite phase noise sound wave at the same time. Then the algorithm is modified to add the speech signal to the system. Though the performance of the algorithm to reduce the noise in the reverberant condition is worse than in the non-reverberant condition, it is also contributive to improve the clarity of the received speech signal with the ANC system by reducing the background noise in the environment. The evaluation index of the simulation results demonstrates this conclusion. Secondly, we study on the inverse filter to improve the speech signal clarity by eliminating the speech reverberation caused by the propagation path. The inverse filter of the propagation path was calculated based on the fast inverse filter (FIF) algorithm. The original speech signal was preprocessed by the inverse filter to make the received speech signal with less reverberation through the propagation path. The propagation model was run to verify the correctness of the inverse filter algorithm. The simulation results indicate that the speech signal preprocessed by the inverse filter can be improved the speech signal clarity significantly. The same conclusion can be obtained by human subjective feeling. At last, the ANC-IF method that combined the ANC system and the inverse filter was proposed and verified to reduce two types of noise simultaneously in the simulation model. The ANC speaker generated the signal that included both the opposite phase noise signal and the speech signal preprocessed by the inverse filter. Consequently, the speech signal would be received with less background noise and less speech reverberation. The simulation results indicate that the ANC-IF method can control the background noise effectively and reduce the speech reverberation significantly. Evaluation indexes also show the effectiveness of the proposed method.