CAJ | 학술논문

为使用户可以通过哼唱音乐片段的方式，在较大规模乐曲库中能实时、准确、快速搜索到目标歌曲，结合多种音乐的乐理特点和音乐的数字化表示，提出考虑周期性测度改进传统端点检测方案，并综合线性预测分析（LPC）和相关函数与削波法相结合的方法进行旋律基音提取。实验表明利用LPC仅仅保留声门激励信号的特点，有效地减少了声道共振峰的影响，而采用削波法也一定程度减少噪声的干扰并使得计算相关函数的运算量大为减少，更显著的是考虑周期性测度的端点检测可以找到更精确的语音段，可以准确提取旋律并有效达到哼唱检索实时性要求。
위사용호가이통과형창음악편단적방식，재교대규모악곡고중능실시、준학、쾌속수색도목표가곡，결합다충음악적악리특점화음악적수자화표시，제출고필주기성측도개진전통단점검측방안，병종합선성예측분석（LPC）화상관함수여삭파법상결합적방법진행선률기음제취。실험표명이용LPC부부보류성문격려신호적특점，유효지감소료성도공진봉적영향，이채용삭파법야일정정도감소조성적간우병사득계산상관함수적운산량대위감소，경현저적시고필주기성측도적단점검측가이조도경정학적어음단，가이준학제취선률병유효체도형창검색실시성요구。
For searching the target song in real-time, accurately and quickly in a rather large music library by humming music clips, integrating music theory characteristics and digitized representation of a variety of music, this paper presents a method of melody pitch extraction by the period measure improving traditional endpoint detection and integrated approach of Linear Predictive Coding(LPC)and correlation function with clipping method. Experiments show that LPC can reserve glottal excitation signal characteristics and eliminate the influence of the sound tract impulse response and clipping method can reduce the noise interference and cut down calculated amount dramatically. More significantly, the method increases the veracity of extracting music clips by period measure improving traditional endpoint detection. As a result, the method meets the requirements of accuracy and real-time for extracting melody and target song.