基于自相关函数的钢琴乐音改进识别算法

Improved Piano Audio Recognition Algorithm Based on Autocorrelation Function

在传统三电平削波结合自相关函数识别算法的基础上,经过准确的音频分割后,提出了帧移法提取乐音基音信号。该算法能在更精细尺度上寻找最大自相关函数,进而准确定位基音位置,较好地解决了传统算法中当乐音节奏较快时,无法区分半频和倍频对基音的影响,从而导致的识别率低的问题。实验表明,本算法对于节奏快慢不同的钢琴乐音的平均识别率约为83.0%,且快节奏乐音的识别率较传统算法高出20.2%,因此该方法对乐音识别尤其对快节奏乐音识别有显著效果。

Combined with traditional three-level center clipping method and autocorrelation function recognition algorithm, an improved frame-shift algorithm to extract precisely the pitch signal was presented, which could search the maximum autocorrelation function at a finer scale to accurately locate the pitch position after accurate audio segmentation. This algorithm solved the problem that the traditional algorithms could not distinguish the influence of half-frequency and double-frequency on the pitch with fast rhythm, which degraded the recognition rate. Experiments showed that the improved algorithm had an average recognition rate of 83.0% for piano music with different rhythms, and the recognition rate with fast-paced music was 20.2% higher than that of traditional method. Therefore, the proposed algorithm has a significant improvement on music recognition, especially for fast-tempo music.