面向视唱评价的声乐自动标注系统

发布时间：2018-02-25 16:36

  本文关键词： 音高提取 音符切分 音乐自动转写 视唱评价　出处：《清华大学学报(自然科学版)》2011年12期 　论文类型：期刊论文

【摘要】：歌唱声音的自动标注是基于内容的音乐分析和检索领域的基础问题。该文在统计分析汉语歌唱声音声韵母时长和音高特征的基础上,提出了一种声乐自动标注模型。该模型将信号处理、语音学和音乐知识结合,通过韵母-乐谱音高对齐和基于声母时长分布的边界优化算法实现了高精度的音符切分。音高提取算法在移调检测的基础上,通过限定基音周期搜索范围的策略克服了信号频谱中半频/倍频点的干扰。结合音乐教学实践,通过检测音符内部最稳定音高成分、估计演唱速率和归一化音符时长的方法提取歌唱声音的音高、节奏和时长信息,并基于这3个客观物理量提出了一个视唱评价方法。实验表明,该文提出的算法能够准确地标注歌唱声音中的音符边界(平均误差26ms)和音高(平均误差0.23半音)。
[Abstract]:The automatic tagging of singing sounds is the basic problem in the field of content-based music analysis and retrieval. A vocal automatic tagging model is proposed, which combines signal processing, phonetics and music knowledge. High precision note segmentation is realized by using vowel-music pitch alignment and boundary optimization algorithm based on initial time distribution. Pitch extraction algorithm is based on shift detection. By limiting the search range of pitch period, the interference of half frequency / double frequency point in the signal spectrum is overcome. In combination with music teaching practice, the most stable pitch components within the notes are detected. The method of estimating singing rate and normalized note duration is used to extract the pitch, rhythm and duration information of singing sound. Based on these three objective physical quantities, a method for evaluating the singing performance is proposed. The experimental results show that, The algorithm proposed in this paper can accurately label the note boundary (average error 26ms) and pitch (average error 0.23 semitones) in singing sound.
【作者单位】： 清华大学计算机科学与技术系;西北师范大学物理与电子工程学院;
【基金】：国家自然科学基金资助项目(60928005,60910130)
【分类号】：J619;TN912.3
，

本文编号：1534339