Skip to content
apleGitHubPublic

About

a vietnamese corpus for diffsinger

Resources

Stars

1 star

Watchers

0 watching

Forks

Latest commit

 

History

78 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 

Repository files navigation

Logo

a vietnamese corpus for diffsinger

yes i did make a logo. god forbid a girl has hobbies!


[ introduction ]

NOTE: i am not fluent in vietnamese (sorry ancestors) so pronunciation may seem off. this corpus serves as a way to cover all the phonemes and shouldn't be used as the only vietnamese corpus in your dataset (if there are any that are public to begin with). let's hope it isn't too scuffed! hopefully this can allow other public vietnamese corpuses to be created, i don't want to be the only one which in turn can fuck up a lot of pronunciations!

first diffsinger dataset heyyyy! i decided to make a pretty scuffed corpus for vietnamese support! i mainly recorded this since reclists are lowkey the best way to cover every phoneme under the bus but anygays! this is extremely scuffed but i hope this can be of help!

[ dataset info ]

477 .wav files

28 minutes and 7 seconds

denoised, dereverbed, and normalized

recorded at 16 bit 44.1k hz mono

labeled in .lab format

uses vi as its language code

[ dictionary ]

oh lord

based off of jani tran's dictionary

uses mostly the central dialect

[ phonemizer ]

not made yet tee hee

use lotte's phonemizer: [link]

[ phonemes ]

UPDATE: i have included all missing phonemes (yes including r) since i one day plan to update the dataset to support all dialects. i got no time to record, though, so i am just leaving this out here in case anyone wants to record using this phoneme list.

examples and non-vietnamese approximation have been referenced from wikipedia. most of the non-viet. approximation are in english, but some sounds from different languages have been included as well. most of the phonemes are based off of the central dialect

non vocal phonemes

use phoneme description
silent stops SP pure silence
breaths AP breaths
end breaths exh end breaths
glottal stops Q abrupt stops
vocal fry vf vocal fry

initial consonant phonemes

señorita and amigo are spanish / gnocchi is italian / жена is russian / 肉 is mandarin chinese

ipa phoneme example non-viet. approximation
m m mai my
n n nam no
ɲ nh nhà canyon, señorita, gnocchi
ŋ ng ngâm, nghe singer
p p pin sport
t t tây stop
tʰ th thầy top, taste
tʂ tr tra trend, but with tongue curled
tɕ ch chè change
k k cô, kem scan
ɓ b ba bee with a gulp
ɗ d đi day with a gulp
f ph phở fight, photo
s x xa so
ʂ s sáu show, but with tongue curled
x kh khô loch
h h hàng high
v v về victory
z z da, danh zero
ɣ g ga, ghế amigo
l l là low
j y già, giết yes
ʐ r ra, rồi жена / žená, 肉 / ròu
w w oanh, quốc quick

final consonant phonemes

note: all of the non-nasal and non-semivowel ending phonemes will have no release

ipa phoneme example non-viet. approximation
k̟̚ CH cách technical
k̚ K ác, học pick
ɲ NH bình onion
ŋ NG trứng, chúng long
n N ban, mìn, bên, bốn, bún pin, hen, pen
m M thêm pom
p̚ P tiếp clasp, but no audible release
t̚ T xuất, chít, một pit, hit, cut
j Y cái, tay, tui boy
w W tao, triệu, đau how

vowel phonemes

note: ax is mainly used for diphthongs, it should only be next to i, eu, and u

노래, 음식, and 눈 are korean / ты is russian / 空気 is japanese / 安, 餓, and 土 are mandarin chinese

ipa phoneme example non-viet. approximation
a a ba, mai, cao 安 / ān
ʌ ah ân gut
ɤ oe bơ 餓 / è
e e về, cây day, said (monophthongal)
ɛ eh xe led
i i khi, tuy seat
o o cô, sâu 노래 / norae, story
ɔ ao có, xoong off
ɯ eu tư ты, 음식 / eumsik, 空気 / kūki
u u ru, tui rule, 눈 / nun, 土 / tǔ
ə ax yêu, ước, ưa, uông rime for ê, ớ, a, and ô

[ terms of use ]

- you are allowed to use this dataset for multispeaker training of svs models

- you are allowed to use this dataset for research purposes

- you are allowed to use this dataset for training commercial svs models

- you are allowed to use this dataset to train an svs model of this voice, but you are not allowed to release it


- you are not allowed to redistribute the raw files of this dataset

- you are not allowed to use this dataset for multispeaker training if your dataset contains vocalists that haven't given you permission for ai training

- you are not allowed to run this dataset through any svc technology to morph it into other vocals

- you are not allowed to use this dataset to train an svc model

[ credits ]

aple: creator, voice provider, labeling

alex floarea: name provider, labeling

[ license ]

ViCa © 2026 by aple is licensed under CC BY 4.0

About

a vietnamese corpus for diffsinger

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors