Repository navigation
BETA DiffSinger Japanese Extended Phonemizer (all versions)
·
8 commits
to master
since this release
A rule-based phonemizer that reduces the need for an insanely long dsdict-ja.yaml. Can handle diphthongs, sokuon, and more, requiring only phoneme definitions. Now using one release page for all variants.
If you use OpenUTAU(the main version):
- use the file marked MAIN
- built for .NET 10 and as such requires the beta version at time of writing
- uses DiffSingerG2pPhonemizer class, which does not support one-to-many replacements and locks out some of the full potential
- appears in the Japanese category of the phonemizer list as DIFFS JA+
If you use OpenUTAU Lunai Edition:
- use the file marked LUNAI
- built for .NET 8
- uses DiffSingerRefinedPhonemizer class, and so does support one-to-many replacements for the more obscure phonemes.
- appears in the phonemizer list as Japanese+
If you use UTAUV:
- use the file marked OTHER
- functionally identical to the version for Lunai, but with the naming conventions of the main version
GENERAL USAGE NOTES:
- In a majority of situations, this will be plug and play.
- Some extremely obscure phonemes may not be available for most models. See DIFFS_JA+_phoneme_notes.md for details.
- cl is always treated as a global phoneme. Models with ja/cl instead will need to add - {from: cl, to: ja/cl} to the replacements section of dsdict-ja.yaml.
v0.1.3 changes: better handling of things like いぇ and うぃ