低信噪比下公共场所异常声音声学特征提取

为了实现低信噪比下公共场所异常声音声学特征提取,提出经验小波滤波器组用于提取异常声音声学特征。首先,根据等效矩形带宽的人耳听觉特性,得到各滤波器的中心频率,计算出经验小波滤波器组的边界。然后,将边界代入经验小波细节函数和尺度函数中,形成经验小波滤波器组。最后,用经验小波滤波器组分解低信噪比下公共场所异常声音,经分解的各模态归一化对数能量作为异常声音声学特征,用于分类识别。相关实验表明,提出的经验小波滤波器组与典型的语音信号处理及时频信号处理方法相比,在低信噪比(0 dB)的商店、银行、办公室、自动取款机环境下,对异常声音的平均识别率提高了4.75%~37.92%,验证了提出方法的有效性。 

Acoustic features extraction of abnormal sounds in public places with low signal-to-noise ratio

In order to obtain the acoustic features of abnormal sounds in public places with low signal-to-noise ratio,we propose an empirical wavelet filter bank for extracting acoustic features of abnormal sounds.First,the center frequencies of the filters of the filter bank are obtained based on the human ear hearing characteristics of the equivalent rectangular bandwidth.The boundaries of the empirical wavelet filter bank are calculated.Then,the boundaries are introduced into the empirical wavelet scale function and scaling function to form the empirical wavelet filter bank.At last,an empirical wavelet filter bank is used to decompose the abnormal noise in public places under low signal-to-noise ratio.The normalized logarithmic energy of each decomposed modal is used as the abnormal acoustic acoustics for classification and recognition.The experiments show that compared with the typical speech signal processing and time-frequency signal processing methods,the empirical wavelet filter proposed in this study improves the average recognition rate by 4.75%~37.92%in 0 dB SNR shops,banks,offices and ATMs the validity of this method is confirmed.