Skip to content
maeblythe edited this page Sep 12, 2024 · 2 revisions

Phonemes

While labeling and using your Diffsinger model, phonemes are undoubtedly one of the most important things.

It is advised to only use ASCII characters and avoid special characters like "{" and "@" within phoneme names. However, other than + and -, there are no restrictions on what characters or combination of characters (excluding any including spaces) can be used at the time of writing.

However, there are already several phonemizers for Diffsinger in existence. Using a novel system in which every phoneme is represented by a number would mean that every USTx you make in OpenUtau would require changing the name of every note.

Using special characters as phonemes

Phenetic systems like X-SAMPA contain a lot of characters that will break your dsdict file such as "@" and "{".

You can escape these special characters in the YAML dsdict file by surrounding the special character with quotation marks. Then, your model will work perfectly with those characters.

Remapping Phonemes using the Dict File

If possible, please add to this section explaining if it is feasible to remap a model so that it can work with default phonemizers. Especially if it can be set to work differently with different phonemizers.

Existing Phonetic Systems

While you can make your own system and phonemizers, it is advisable to stick with what is known to work. The following are the phonemes used for each phonemizer and where to download it if it does not come default with OpenUtau.

Phonemes outside of the scope of each phonemizer can be added to add extra language capabilities without issues.

If a phoneme used by the phonemizer is not within the model, an error will occur.

English

Download: OpenUtau default

Phonemes: Arpabet Phonemes

Japanese

Download: OpenUtau Default

Phonemes: Synth V Japanese Phonemes