Variance embeds in multispeaker models

[ariikamusic]

Acknowledgement

• I have read Getting-Started and FAQ

🐛 Describe the bug

I am writing this issue on behalf of multiple community members (Discord) who are or have been experiencing the issue.

Whenever variance (pitch, tension, voicing) in DiffSinger is generated for multispeaker models, OpenUTAU seems to generate the variance model of only the first speaker embed inside the configuration file, regardless of choosing the specific speaker inside OpenUTAU. It is mentioned that the variance model and/or generated pitch stays very similar if not the same with all speakers.

Explains how to reproduce the bug

N/A

OS & Version

Windows

Logs

N/A