百合と蜂 — Japanese synthetic dialogue, version 0.8

232 active Japanese dialogue recordings with Korean subtitles.
Engine: Qwen3-TTS 12Hz 1.7B VoiceDesign + Base, qwen-tts 0.1.1.
Model/project license: Apache-2.0.
https://github.com/QwenLM/Qwen3-TTS
https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign
https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-Base

Hachisuka, Mashiro, Sayuri, Risa and Popo each use an original synthetic
voice designed with text instructions. No real actor recording was used
as a cloning reference. These are AI voices, not human performances.

Generation runs on the server. The game contains only finished 24 kHz mono
OGG recordings; it does not load a speech model on the player's computer.

Naturalness is subjective. Japanese ASR was used to check transcription
alignment, separately from waveform/coverage and in-game playback tests.
ASR is not a native-speaker accent or acting review.

Version 0.6: 20 marked voice slots in added/replaced scenes (19 newly
synthesized, 1 reused spoken line), including one chat voice message.
Unchanged legacy dialogue retains its existing files. New non-star lines are
unvoiced.
New crying-bee and teacup stickers: OpenAI ImageGen, original generated assets.

Version 0.7: 115 previously unvoiced NPC dialogue slots completed.
195 NPC recordings plus 37 retained optional player recordings. Player voice is
muted by default. Written chat and narration remain unvoiced.
12 separate A/B/C auditions include AivisSpeech まお and コハク.
See TTS_AUDITION_CREDITS.txt for the full ACML terms and model attribution.

Version 0.8 recast (Sayuri, Risa, Popo; 125 recordings regenerated).
Sayuri and Risa: new Qwen3-TTS VoiceDesign voices written from text
descriptions (adult 25-year-old / 28-year-old), cloned with Qwen3-TTS Base.
No human recording was used as a reference.
Popo: AivisSpeech: まお (卯畑まお ©Oz Chat/Trippy, CV: ねゆたろ),
model a59cb814-0083-4369-8542-f51a29e72af7, ACML 1.0. Styles からかい / せつなめ.
https://hub.aivis-project.com/aivm-models/a59cb814-0083-4369-8542-f51a29e72af7
Mashiro and Hachisuka recordings are unchanged from version 0.7.
