Human-voice-only dataset including audio drama CD

Fairy 02, 発艦！

Voices from anime and audio drama are both included. Music is removed via demucs.

Please create pull requests or issues, and fix if you find error in meta.csv or drama-cd-transcript.csv or the srt subtitles in drama-cd-transcript directory. I am not so good at listening to Japanese.

Although I extracted the vocals (removing music), there is no good way to remove the background singing yet. You can try using AI to separate the speakers, or directly extract the singing from the OST and subtract it.

See the previous release https://github.com/Hecate2/sukasuka-vocal-dataset-builder/releases/tag/0.0.anime for more information.

Feel free to edit, repost and re-distribute this dataset. Looking forward to someone building a bittorrent, posting and seeding it publicly.

With images.7z in this release, you can build videos with make_mkv.sh in the repository. ffmpeg may take 13 GB memory.

Character name changes:
Lilya -> Lillia
See character.csv for more information about translations.

Sum of voice time is provided below, grouped by character:

Character	Count	Seconds	Duration (hh:mm:ss.cc)
Willem	1,244	2,684.46	00:44:44.46
Chtholly	1,095	2,408.96	00:40:08.96
Ithea	549	1,403.71	00:23:23.71
Nygglatho	447	1,200.80	00:20:00.80
Tiat	272	739.72	00:12:19.72
(empty)	277	531.35	00:08:51.35
Lillia	167	446.14	00:07:26.14
Nephren	237	432.09	00:07:12.09
Rhantolk	141	385.45	00:06:25.45
Grick	131	321.24	00:05:21.24
Suowong	89	277.70	00:04:37.70
Nopht	116	268.22	00:04:28.22
Pannibal	103	256.49	00:04:16.49
Lakhesh	88	234.93	00:03:54.93
Collon	95	212.52	00:03:32.52
Elq	90	212.50	00:03:32.50
Phyracorlybia	84	208.60	00:03:28.60
Almaria	74	182.25	00:03:02.25
Limeskin	49	129.34	00:02:09.34
SuowongYoung	57	119.73	00:01:59.73
EbonCandle	31	94.04	00:01:34.04
Margomedari	27	92.43	00:01:32.43
Godley	27	83.67	00:01:23.67
Tuka	19	54.42	00:00:54.42
Buronny	12	32.63	00:00:32.63
Almita	18	24.03	00:00:24.03
Narration	2	9.39	00:00:09.39
SilverClover	2	5.72	00:00:05.72

Drama CD transcribed by AI without character labels

Transcribed all the drama CDs with openai-whisper.
Neither the content nor the time is 100% accurate.
Need more manual labelling.

Anime dataset full release + experimental voice synthesis

NOTICE: Go to https://huggingface.co/datasets/mio/sukasuka-anime-vocal-dataset if you want the dataset without non-vocal sounds.
IMPORTANT NOTICE: For now it is always recommended to clone the latest codes along with meta.csv and build the dataset on your own, because meta.csv is frequently changing.

Almaria 28
Almita 10
Buronny 12
Chtholly 764
Collon 31
EbonCandle 31
Elq 90
Godley 27
Grick 131
Ithea 224
Lakhesh 38
Lilya 20
Limeskin 49
Margomedari 27
Narration 2
Nephren 113
Nopht 116
Nygglatho 278
Pannibal 24
Phyracorlybia 84
Rhantolk 141
SilverClover 2
Suowong 89
SuowongYoung 5
Tiat 102
Willem 1027

TOTAL: 3465

Subtitle fixes:
第二集[16分35秒95 到 16分37秒36] , 然の身の上話 应改为 突然の身の上話
第4集[04分24秒95到04分27秒83]，日语字幕是空白的。实际缇亚忒说了 ふおーっ！コリナディルーチェ！
[07集-0261][20.04.51-20.08.77].ogg,Nygglatho,「地上調査隊が獣に襲われた」って　商会から連絡が入ったわ 开头多了个空格
[08集-0047][04.14.31-04.15.52] おん… おん？ 原本字幕是空的

【The source codes of so-vits-svc has been deleted by the owner innnky. Therefore, all the models mentioned in this release are no longer maintained or recommended!】
【successor: https://github.com/svc-develop-team/so-vits-svc 】

Voice model of Chtholly by Aya. The config is also included in this release as backup.
https://drive.google.com/drive/folders/1YzC6XAtXiTsAIzrzrQGhybBFZnGwhf7j

Chtholly-singing-Dearest-Drop.mp3 , synthesized by Aya, using so-vits-svc 4.0 is included in the release as a quick test. Somehow the model is overfitting?
Chtholly being Symboli Rudolf by Aya, using so-vits-svc 4.0

Labelled episodes 01 till 06

Almaria 28
Almita 5
Buronny 12 (バロニ・マキシ, Baroni Makushi, officer of Guardian Wings)
Chtholly 361
Collon 22
EbonCandle 31 (黒燭公)
Elq 38
Grick 35
Ithea 131
Lakhesh 19
Lillia 6
Limeskin 37
Margomedari 27 (one-eyed doctor)
Narration 2
Nephren 54
Nygglatho 158
Pannibal 17
Phyracorlybia 84 (フィラコルリビア・ドリオ, wolf-like, Corna di Luce, Mayor's Daughter)
SilverClover 1 (銀詰草)
Suowong 83 (the great sage)
SuowongYoung 5
Tiat 91
Willem 594

TOTAL: 1841

raw dataset without clustering by character

0.0.raw

voice extractor