基于移频和自适应陷波器的啸叫抑制算法

Howling Suppression Algorithm Based on Frequency Shift and Adaptive Notch Filter

会议扩声场景下,麦克风距离扬声器过近、发言人声音过大、调音台音量调整不合适,均可能引发啸叫现象,导致糟糕的会议体验,甚至造成设备损坏和听力的损伤。对此,提出一种结合移频和自适应陷波算法的啸叫抑制方案,针对上述问题,将麦克风采集的信号先进行微量移频,破坏啸叫的产生条件,再将信号划分8个子频带,每个子频带提取频谱峰度特征判断是否啸叫,对啸叫频点做陷波处理,最后通过增益补偿弥补损失的声压,达到抑制啸叫、提高扬声器增益的目的。在实际的会议扩声测试中,所提的算法可以提供10~18 dB附加增益,有效预防反馈发生。

In the conference sound reinforcement scenario, if the microphone is too close to the speaker, the speaker’s voice is too loud,or the volume of the mixer is not properly adjusted, it may cause howling, resulting in a poor conference experience, and even damage to equipment and hearing. In this paper, a howling suppression scheme combining frequency shifting and adaptive notch algorithm is proposed. In view of the above problems, the microphone collected signal is first subjected to a slight frequency shift to destroy the generating conditions of howling, and then the signal is divided into 8 sub-bands. Each sub-band extracts the spectral kurtosis feature to determine whether or not howling, performs notch processing on the howling frequency point, and finally compensates for the lost sound pressure through gain compensation, so as to suppress howling and improve the speaker gain. In the actual conference sound reinforcement test, the algorithm in this paper can provide an additional gain of 10~18 dB, which can effectively prevent feedback from occurring, which depending on the acoustic environment.