search

sevenc-nanashi / Cantari

UTAUの歌声をVoicevoxで話させるためのエンジン。
https://www.nicovideo.jp/watch/sm43856969
MIT License
10 stars 0 forks source link

voice caching issue #1

Open nagotown opened 4 months ago

nagotown commented 4 months ago

when rendering with the same phonemes after changing voicebank, the cache doesn't clear and the previous render is used. this can result in multiple voices being used depending on which voicebank used which phoneme first.

os: windows 10 22H2 voicevox version: 0.19.2 cantari version: 0.0.6 - 0.0.2

https://github.com/sevenc-nanashi/Cantari/assets/35883871/3be8c86f-df2f-4f4f-a79e-d62e80fa94b1

sevenc-nanashi commented 4 months ago

iirc voicevox editor caches the rendering result. Maybe that's the issue? (Sorry but I didn't understand what the issue is)

nagotown commented 4 months ago

rendered "こんばんは" with mujika (無地歌) after switching the voicebank to teto, it still outputs mujika's rendered phonemes. any phonemes not already rendered will be initialized by the currently selected voicebank

nagotown commented 4 months ago

without restart after changing voicebank:

https://github.com/sevenc-nanashi/Cantari/assets/35883871/d32fc235-2378-4bc0-af6d-f313dca7b594

with restart:

https://github.com/sevenc-nanashi/Cantari/assets/35883871/fba70cd5-d579-438d-a8c4-367c4c41fda9