jaCappella corpus : Japanese a cappella vocal ensemble corpus

The jaCappella corpus is a corpus of Japanese a cappella vocal ensembles. It consists of musical scores of vocal ensemble songs and separate audio recordings of individual voice parts. They were arranged from out-of-copyright Japanese children's songs. All songs have six voice parts (lead vocal, soprano, alto, tenor, bass, and vocal percussion).

Trailer video

Getting started

Download the data in the jaCappella directory of your computer by huggingface-cli

huggingface-cli download jaCappella/jaCappella --repo-type dataset --local-dir jaCappella --local-dir-use-symlinks False

Subsets

The songs of the jaCappella corpus are divided into the following seven subsets.

Subset	Total duration [s]	Songs	Remarks
Jazz	226.7	otamajakushi, popura, nanatsunoko, akaikutsu, machibouke
Punk rock	310.7	anomachikonomachi, kaerutokumo, shabondama, hanyuunoyado, chouchou
Bossa nova	334.5	dongurikorokoro, urashimatarou, ryousen, haruyokoi, touriyanse
Popular	352.5	koinobori, kutsuganaru, akatonbo, yuki, chatsumi
Reggae	228.7	usagitokame, omochanomaachi, kisha, shoujoujinotanukibayashi, muranokajiya
Enka	361.1	hiraitahiraita, fujinoyama, aogebatoutoshi, koujounotsuki, juugoyaotsukisan
Neutral	260.1	katatsumuri, komoriuta, hato, harugakita, momotarou
Ballad	403.5	doubutsuen, hiyoko, kakashi, nakayoshikomichi, shikararete	Added in v2
EDM	391.7	koganemushi,meemeekoyagi,momiji,sunayama,yamaderanooshousan,yamaderanooshousan	Added in v2
Soul/funk	340.9	akaiboushishiroiboushi,dokokadeharuga,inu,makibanoasa,yurikagonouta,yurikagonouta	Added in v2

Each subset includes five songs.

Audio files of singing voices

The singing voices were recorded in a recording studio. The sampling frequency is 48 kHz and the audio file is provided in a monaural WAVE format. All singers are native Japanese speakers.

Score

The musical scores with Japanese lyrics are provided in the PDF and MusicXML formats. This dataset contains MusicXML files whose lyrics were modifed for singing voice synthesis (e.g. "は"->"わ", "へ"->"え"). For non-Japanese researchers, MusicXML files with Romaji transcription of lyrics are provided.

Remark: Romaji transcription does not exactly match Japanese pronunciation. For the exact Japanese pronounciation, please refer to the singing voices.

Metadata

Metadata of the songs are given as a csv file. The items in the csv file are as follows.

title_in_en: Title in Hepburnian Roman alphabet
title_in_ja: Title in Japanese
lyric_writer: Lyric writer name of the original song in Japanese
copyright_of_lyric_writer: Copyright of the lyric writer of the original song
composer: Composer name of the original song in Japanese
copyright_of_composer: Copyright of the composer of the original song
arranger_in_en：Arranger name of the song of our corpus in English
arranger_in_ja：Arranger name of the song of our corpus in Japanese
subset：Subset name
voice_part：Voice part name
singer_id: Singer identifier (ID)
gender: Singer's gender
first_lang: First language of singer

Directory structure

{subset}, {title_in_en}, and {voice_part} refer to the subset name, song title, and voice name, respectively. They are written in lowercase, with spaces replaced by underscores _.

Audio file of voice part: jaCappella/{subset}/{title_in_en}/{voice_part}.wav.
Mixture of all voice parts
- Monaural: jaCappella/{subset}/{title_in_en}/mixture.wav
- Stereo: jaCappella/{subset}/{title_in_en}/mixture_stereo.wav
Musical score
- PDF with Japanese lyrics: jaCappella/{subset}/{title_in_en}/{title_in_en}.pdf
- MusicXML with Japanese lyrics: jaCappella/{subset}/{title_in_en}/{title_in_en}.musicxml
- MusicXML for singing voice synthesis (SVS) with Japanese lyrics: jaCappella/{subset}/{title_in_en}/{title_in_en}_SVS.musicxml
  - Nonlexical syllables are capitalized (e.g., LA, TU).
- MusicXML with Romaji transcription of lyrics: jaCappella/{subset}/{title_in_en}/{title_in_en}_romaji.musicxml
  - Nonlexical syllables are capitalized (e.g., LA, TU).
  - Lexical syllables are not capitzlied (e.g., cha, i).
Metadata (CSV): jaCappella/meta.csv

jaCappella
├─meta.csv
├─test_song_list_for_vocal_ensemble_separation.txt # This file includes song titles used as test data in vocal ensemble separation of our paper.
├─ballad
├─bossa_nova
├─edm
├─enka
├─jazz
├─neutral
│  ├─katatsumuri
│  │  ├─katatsumuri.pdf
│  │  ├─katatsumuri.musicxml
│  │  ├─katatsumuri_SVS.musicxml
│  │  ├─katatsumuri_romaji.musicxml
│  │  ├─lead_vocal.wav
│  │  ├─soprano.wav
│  │  ├─alto.wav
│  │  ├─tenor.wav
│  │  ├─bass.wav
│  │  ├─vocal_percussion.wav
│  │  ├─mixture.wav
│  │  └─mixture_stereo.wav
│  ├─komoriuta
│  ├─hato
│  ├─harugakita
│  └─momotarou
├─popular
├─punk_rock
├─reggae
└─soulfunk

Terms of Use

Please refer to the jaCappella Corpus website for the terms of use.

Contributors (The jaCappella corpus team)

Tomohiko Nakamura
Shinnosuke Takamichi (The University of Tokyo)
Naoko Tanji (The University of Tokyo)
Hiroshi Saruwatari (The University of Tokyo)

Citation

If you use the jaCappella corpus, please cite our paper. Please see the jaCappella webpage.

Acknowledgement

The jaCappella corpus team reserves all copyrights and neighboring rights of the vocal ensemble songs. The music arrangement and sound production were performed by Alpha Enterprises, Inc.

Change logs

Dec. 2023: v.2.0.1
- Fixed bug: Modified bass.wav and vocal_percussion.wav of yurikagonouta in the soul/funk subset
Dec. 2023: v2
- Added sheet music and audio recordings of 15 songs (ballad, EDM, and soul/funk)
Feb. 2023: v1.1
- Some accidental marks modified in accordance with harmony.
- Added MusicXML files for SVS
- Added MusicXML files with Romaji transcriptions of lyrics

Known issues

Audio

Title	Voice part	Location	Content
kaerutokumo	soprano	Measure 31, 2nd note	sung with not "ha" but "ah"
fujinoyama	bass	Measure 5, 3rd note	sung with not "du" but "da"
juugoyaotsukisan	lead_vocal	Measure 15, 1st note	sung with not "wa" but "ah" in second verse

Datasets:

jaCappella
/

jaCappella

You need to agree to share your contact information to access this dataset