CAJ | 학술논문

为了提高合成歌声的自然度.分析了大规模视唱语料库中的歌唱声音的颤音的声学特征,并应用分析所得的颤音特点辅助歌声合成,实验结果表明,合成的歌曲达到了3.52的平均MOS得分.另外,提出了一种基于希尔伯特黄变换(Hilbert-Huang Transform,HHT)的颤音检测算法,此算法用于指导视唱评价过程可提高评价模型的准确度.
위료제고합성가성적자연도.분석료대규모시창어료고중적가창성음적전음적성학특정,병응용분석소득적전음특점보조가성합성,실험결과표명,합성적가곡체도료3.52적평균MOS득분.령외,제출료일충기우희이백특황변환(Hilbert-Huang Transform,HHT)적전음검측산법,차산법용우지도시창평개과정가제고평개모형적준학도.
@@@@In order to improve the naturalness of synthesize singing voice. The characteristics of vibrato in large-scale sight-sing-ing corpus of are analyzed firstly, and the analysis results are used to assist the singing voice synthesis. Experiments demonstrate that it can achieve 3.52 of average MOS score. Then, a vibrato detection algorithm based on HHT is presented which may guide the evaluation process and improve the accuracy of the evaluation model.