基于线性预测的综合基音检测法

Integrated pitch detection method based on linear predictive analysis

由于传统基音检测方法在信噪比较低时对弱浊音和清浊音过渡帧的检测效果较差,提出了一种综合的基音检测法。该方法利用线性预测分析和低通滤波器对含噪语音进行预处理,再采用改进的平均幅度差函数加权自相关函数的方法对浊音帧进行基音检测,对于清浊音过渡帧,则利用小波变换进行基音检测。实验结果表明,该法提高了检测精度,即使信噪比降至-5 dB,仍能较准确地检测大部分弱浊音和清浊音过渡帧的基音周期,准确性和鲁棒性较好。

An integrated pitch detection method is proposed since the traditional methods can＇t get good performance for weak sonant frames and transition frames ofsurd and sonant in low signal to noise ratio, In the method, linear predictive analysis and low pass filter is used to preprocess with the noisy speech. Then, pitch periods of sonant frames are detected adopting the method of autocorrelation function weighted by the modified average magnitude difference function. As for the transition frames ofsurd and sonant, pitch periods are detected using wavelet transform, The experimental result shown that the new method improves the precision of detection, It can detect pitch periods of most weak sonant frames and transition frames of surd and sonant even when the signal to noise ratio is as low as -5 decibels, so it has good accuracy and robustness.