卡拉OK歌唱評分方法研究

A Singing Evaluation Method for Karaoke Applications

黎欣捷(Hsin-Chieh Lee)

指導教授 ： 蔡偉和

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2010年)

https://doi.org/10.6841/NTUT.2010.00174

摘要

本研究的目標在於模擬卡拉OK 歌唱環境下，利用卡拉OK VCD 所含音樂資
料進行演唱者的歌唱能力好壞分析，所發展的歌唱評分系統具備三項評分機制，
分別依據「音高」、「動態音量」以及「和諧度」進行分析，使演唱者了解其歌
唱優劣之處為何，以及應如何改善。其中「音高」反應演唱音符是否正確，「動
態音量」代表演唱者詮釋歌曲的情境是否合宜，而「和諧度」則反應演唱者之
節奏是否正確，當歌唱的節奏正確時，歌聲與伴奏所組合的聲音才能和諧悅耳。
本論文特別著力於「和諧度」的分析。雖然估算音樂節奏的方法已存在於許多
文獻，但它們皆只適合用於純樂器的演奏音上，無法用於清唱歌聲中。主要原
因是清唱歌聲並不似純樂器音樂具有明顯的起始點以及強弱拍。針對此問題，
本論文提出一種判斷清唱歌聲與伴奏音樂之間和諧關係的方法，利用和諧與非
和諧音的機率模型統計，進行歌唱節奏好壞評估。另一方面，我們設計一「哼
�唱判別」單元以判斷使用者是否僅哼歌曲而未將歌詞唱出。為了測試系統之
準確性，我們邀請了多位歌唱能力不同的男女參與者錄製清唱歌聲，並聘請若
干人工評分人員進行測試歌聲的人工評分。初步實驗結果顯示，本研究提出之
評分方法與人類感知評分相去不遠，已能有效判別使用者歌唱能力之優劣。

關鍵字

歌唱能力評估 ； 和諧度 ； 節奏 ； 哼/唱判別

並列摘要

This work investigates automated techniques for evaluating the skill of a singer performing with Karaoke VCD music. The proposed system is capable of analyzing the pitch, dynamic volume, and harmonicity in a singing performance, and indicates how well or bad a singer performs. The pitch information reflects whether or not the notes sung by a singer are correct; the dynamic volume conveys the emotion that should be presented in performing a song; the harmonicity reflects whether or not a singing performance is in rhythm. Our special effort in this work is put on harmonicity analysis. It is observed that only when a singing performance is in rhythm, the mix of singing voice and accompaniments can be harmonious and melodious. Although there are copious works studying rhythm extraction from music, all of them only suitable for instrument-solely music rather than singing. For this reason, we develop methods to evaluate the harmonicity between singing voice and accompaniment. The basic strategy is to represent harmonious sound and inharmonious sound as probabilistic models and then perform stochastic recognition. On the other hand, this study also proposes a method for distinguishing between singing and humming, in order to check if a Karaoke user does sing the lyrics rather than simply hums the tune of a song. In our experiments, we invited a number of participants having different singing capabilities to perform solo singing. We also employed several professional judgers to grade each piece of solo singing in our collection. It is found that the grading results obtained with our system are roughly close to those with manual grading, which confirms the validity of our system.

並列關鍵字

Singing Performance Evaluation ； Harmonicity ； Rhythm ； Singing/Humming Discrimination

參考文獻

[13] 許泰寧，數位音頻分析系統研究-以應用於KTV評分，碩士論文，台灣大學，2006。

[1] J. Downie, K. West, A Ehmann, and E. Vincent, “The 2005 music information Retrieval evaluation exchange (MIREX 2005): Preliminary overview,” in Proc International Conference on Music Information Retrieval, London, UK, 2005, pp. 320-323.

[3] W. H. Tsai and H. M. Wang, “Automatic singer recognition of popular music recordings via estimation and modeling of solo vocal signals,” IEEE Trans. on Audio, Speech, and Language Processing, vol. 14, no. 1, pp. 330-341, January 2006.

[4] W. H. Tsai and H. M. Wang, “Automatic identification of the sung language in popular music recordings,” Journal of New Music Research, vol. 36, no. 2, pp. 105-114, 2007.

[5] M. Goto, “A predominant-F0 estimation method for CD recordings: MAP estimation using EM algorithm for adaptive tone models,” in Proc. IEEE Int. Conf. Acoustics, Speech, and Signal Processing, 2001.