Period Singer: Integrating Periodic and Aperiodic Variational Autoencoders for Natural-Sounding End-to-End Singing Voice Synthesis

Taewoo Kim, Choongsang Cho, Young Han Lee
Korea Electronics Technology Institute, South Korea

Abstract


In this paper, we present Period Singer, a novel end-to-end singing voice synthesis (SVS) model that utilizes variational inference for periodic and aperiodic components, aimed at producing natural-sounding waveforms. Recent end-to-end SVS models have demonstrated the capability of synthesizing high-fidelity singing voices. However, owing to deterministic pitch conditioning, they do not fully address the one-to-many problem. To address this problem, we present the Period Singer architecture, which integrates variational autoencoders for the periodic and aperiodic components. Additionally, our methodology eliminates the dependency on an external aligner by estimating the phoneme alignment through a monotonic alignment search within note boundaries. Our empirical evaluations show that Period Singer outperforms existing end-to-end SVS models on Mandarin and Korean datasets. The efficacy of the proposed method was further corroborated by ablation studies.

This page is for research demonstration purposes only.



Model Overview


Fig 1. Overall architecture of Period Singer


Mandarin Chinese Singing Voice Samples


Sample 1: 雨淋湿了天空灰得更讲究 (Yǔ lín shīle tiānkōng huī dé gèng jiǎngjiù)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 2: 这最后的出口在爱过了才有 (Zhè zuìhòu de chūkǒu zài àiguòle cái yǒu)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 3: 这首歌给你快乐你有没有爱上我 (Zhè shǒu gē gěi nǐ kuàilè nǐ yǒu méiyǒu ài shàng wǒ)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 4: 青春有太多未知的猜测成长的烦恼算什么 (Qīngchūn yǒu tài duō wèizhī de cāicè chéngzhǎng de fánnǎo suàn shénme)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 5: 像有着心事的一张表情 (Xiàng yǒuzhe xīnshì de yī zhāng biǎoqíng)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 6: 始终年轻 (Shǐzhōng niánqīng)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 7: 小酒窝长睫毛是你最美的记号 (Xiǎo jiǔwō zhǎng jiémáo shì nǐ zuìměi de jìhào)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 8: 终于找到心有灵犀的美好 (Zhōngyú zhǎodào xīnyǒulíngxī dì měihǎo)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 9: 醒来灿烂星光透过了窗台 (Xǐng lái cànlàn xīngguāng tòuguòle chuāngtái)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 10: 相信我破晓会跨山海准时到来 (Xiāngxìn wǒ pòxiǎo huì kuà shānhǎi zhǔnshí dàolái)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger


Korean Singing Voice Samples


Sample 1: 이제 이별 뒷면에 서있는 우리 (ije ibyeol dwismyeon-e seoissneun uli)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 2: 우리 같이 나눈 시간 그 한 순간도 의미 없는 날은 없었고 (uli gat-i nanun sigan geu han sungando uimi eobsneun nal-eun eobs-eossgo)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 3: 너는 내가 사랑하니까 더 소중하니까 (neoneun naega salanghanikka deo sojunghanikka)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 4: 그저 내 곁에만 있어줘 떠나지 말아줘 (geujeo nae gyeot-eman iss-eojwo tteonaji mal-ajwo)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 5: 내 주머니 속에 넣고 니 손가락 만지작대며 (nae jumeoni sog-e neohgo ni songalag manjijagdaemyeo)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 6: 그럼 우린 다시 사랑스럽게 (geuleom ulin dasi salangseuleobge)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 7: 사계절이 와 그리고 또 떠나 (sagyejeol-i wa geuligo tto tteona)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 8: I gave you the world 너만이 전부라 (I gave you the world neoman-i jeonbula)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 9: 바보 같은 사랑도 괜찮아요 이렇게 그댈 (babo gat-eun salangdo gwaenchanh-ayo ileohge geudael)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger
Sample 10: 내가 그대라면 좋을텐데 그 아픔도 다 가져갈텐데 눈물이 말해 (naega geudaelamyeon joh-eultende geu apeumdo da gajyeogaltende nunmul-i malhae)
Ground Truth VITS VISinger VISinger2 PeriodSinger(DPP) PeriodSinger


Ablation Samples


Sample 1: 완전히 돌아갈 수는 없는 걸 (wanjeonhi dol-agal suneun eobsneun geol)
PeriodSinger w/o NF w/o Note Boundary
Sample 2: 이제 마지막 장을 넘겨야겠어 (ije majimag jang-eul neomgyeoyagess-eo)
PeriodSinger w/o NF w/o Note Boundary
Sample 3: I gave you the world 너만이 전부라 (I gave you the world neoman-i jeonbula)
PeriodSinger w/o NF w/o Note Boundary
Sample 4: 나를 기억하나요 나를 알아보나요 (naleul gieoghanayo naleul al-abonayo)
PeriodSinger w/o NF w/o Note Boundary
Sample 5: 사랑해 그대 부르고 또 불러 봐도 (salanghae geudae buleugo tto bulleo bwado)
PeriodSinger w/o NF w/o Note Boundary