基于卡尔曼滤波的乐音基频小波自相关检测法

An autocorrelation detection method of wavelet for musical pitch based on Kalman filter

针对乐器演奏音域宽广的特点以及加性环境噪声的干扰,传统的基频检测算法无法在整个乐音音域内都能获得精准的基音识别。鉴于此,提出了一种基于卡尔曼滤波和小波变换的乐音基频自相关检测法。该方法利用卡尔曼滤波器滤除全频段大部分噪声,而后对小波变换的各层细节分量阈值进行处理,以提高信号自相关函数波形的准确性和平滑性,从而提高乐音基频检测的精度。实验表明,该方法能够于低信噪比背景下很好地抑制乐音中混杂的白噪声,实现在全频域段内对音频质量的增强以及乐音基频的准确检测,且在全频域段内的平均识别误差率与傅里叶变换对小波变换低通分量的基频检测算法相比下降了0.89%,其鲁棒性更为理想。

Inasmuch as the wide pitch range of musical instruments and the interference of additive environmental noise,the conventional algorithms of pitch detection cannot obtain accurate pitch recognition in the whole musical tone range.In consideration of this,a method of musical pitch autocorrelation detection based on Kalman filter and wavelet transform is proposed.In this method,the accuracy and smoothness of the signal’s autocorrelation function waveform are improved by utilizing Kalman filter combined with the wavelet threshold method for denoising,so as to improve the accuracy of pitch frequency detection of music.Experimental results show that the proposed method can effectively suppress the white noise,enhance the audio quality and detect the pitch frequency accurately in the full frequency domain under the condition of low Signal-to-Noise Ratio(SNR),and it is proved to be a robust method for the fact that the average error rate in the full frequency domain is reduced by 0.89%compared with the pitch detection algorithm of the Fourier transform combined with the wavelet transform method.