一种钢琴全频段的基频识别方法

Approach to Piano Pitch Recognition Covering Full Frequency Range

对钢琴全频段音符频率进行了研究，提出一种基于自相关和插值的钢琴基频识别方法。采用三电平中心削波的自相关计算得出函数值，用求最大值的方法选取代表基频整数倍周期的点数位置，运用抛物线插值进行位置修正，结合最小二乘计算得出基频估计，低中频的识别误差在±1音分内。对于高频音符，采用内插算法提高采样率以降低识别误差，使得识别结果准确可靠，满足精度要求。

To study piano notes covering full frequency range, and an approach for piano pitch recognition is proposed based on auto-correlation and interpolation. It draws function waves after three-level center clipping auto-correlation computing, and chooses the places of fundamental frequency periods by the way of calculating max value. An interpolation method in parabola is used to modify the places, then the fundamental frequency of signal is estimated by the least square method, and the recognition error for signals with LO-MID fundamental is less than -＋ 1 cent. As for signals with high fundamental, it use a method to reduce error based on interpolation for up-sampling. And after that the recognition result is more accurate and reliable, meeting the accuracy requirement effectively.