基於頻譜轉換之自動化歌手辨識與模仿

隨著網路與多媒體設備的日益普及，個人化卡拉OK正逐漸興起。為了更加增其實用性與娛樂價值，本論文探討兩項卡拉OK的附加功能：歌手查詢與歌手模仿。在歌手查詢方面，我們發展一項流行音樂的歌手辨識技術，以降低建置歌手查詢功能時所需人工標示音樂資料的負擔。而在歌手模仿方面，我們開發一套歌聲轉換系統，將使用者的歌聲轉變成另一人的歌聲，達到歌手模仿的娛樂效果。其中，流行歌手辨識的技術關鍵在於如何有效取出含有伴奏的清唱歌聲特性，以建立各個歌手的歌聲模型。本論文提出一種清唱頻譜萃取法，將含有伴奏的歌聲頻譜「轉換」成沒有伴奏的清唱歌聲頻譜。這種「轉換」是建立在自動歸納大量清唱歌聲與疊加伴奏後的頻譜對應關係，將此對應關係表示成轉換參數模型，再利用此參數模型將含有伴奏的未知歌聲頻譜轉換成清唱歌聲頻譜。實驗結果顯示此方法確實可提升流行歌手辨識的準確度。另一方面，我們將頻譜轉換的概念使用於歌手聲音轉換。假設使用者欲模仿某位歌手的歌聲，則系統先透過彼此演唱相同的歌曲片段資料來求取兩者的歌聲轉換關係，之後根據此轉換關係將使用者之任意歌聲頻譜轉換為「目標歌聲」頻譜，再搭配基頻的調整而合成欲模仿歌手的聲音。由主觀測試實驗證實此方法確實可達到某種程度的歌手模仿效果。

關鍵字

歌手識別；歌手模仿；頻譜轉換

並列摘要

Personalized Karaoke is gaining popularity as the wide spread of network and multimedia equipment. To make Karaoke more functional and entertaining, this work studies two added features: singer-based music retrieval and singer imitation. For singer-based music retrieval, we develop a technique for automatically identifying the singers in popular music recordings. This helps users quickly establish the indices of music data according to their associated singers, especially when a new set of songs needs to be included. Since most popular music contains background accompaniments in singing, we focus on the problem of how to extract singer voice characteristics from accompanied singing. Our proposed solution is to “transform” the accompanied singing into solo singing by exploiting the relationships between solo singing and its accompanied versions in spectrum. The relationships are inferred from a large set of solo singing and their accompanied counterparts generated manually. Our experiments show that such a spectrum transform approach increases the accuracy of singer identification noticeably. On the other hand, the spectrum transform is then used to convert singing voices from one person to another. This is done by first finding the transformation of singing spectrum between a user and the person to be intimated, where both the singers perform few the same songs. Then, the transformation, together with pitch shifting, are used to convert any songs performed by the user to the intimated singer. Our experiments based on subjective listening test show that the proposed method achieves singer intimation to some degree.

並列關鍵字

singer identification ； singer imitation ； spectrum conversion

參考文獻

[2]C. C. Liu and C. S. Huang, “A singer identification technique for content-based classification of MP3 music objects,” in Proc. Int. Conf. Information and Knowledge Management, McLean, VA, 2002, pp. 438–445.

[3]T. Zhang, “Automatic singer identification,” in Proc. IEEE Int. Conf. Multimedia Expo, Baltimore, MD, 2003.

[4]M. A. Bartsch and G. H. Wakefield, “Singing voice identification using spectral envelope estimation,” IEEE Trans. Speech Audio Process., vol. 12, no. 2, pp. 100–109, Mar. 2004.

[5]T.L. Nwe and H. Li, “Exploring Vibrato-Motivated Acoustic Features for Singer Identification,” IEEE Trans. Speech Audio Process., vol. 15, no.2, pp. 519-530, 2007.

[6]T.L. Nwe and H. Li, “On fusion of timbre-motivated features for singing voice detection and singer identification,” in Proc. IEEE Int. Conf. Acoustics, Speech, and Signal Processing (ICASSP ), Mar. 2008, pp.2225-2228.

國際替代計量

基於頻譜轉換之自動化歌手辨識與模仿

未授權

主題瀏覽