[IDEA] Improve sentence segmentation with yasbd-lib

这个功能请求是用来解决什么问题的？ / Is your feature request related to a problem? Please describe.

Open-LLM-VTuber uses PySBD for sentence boundary detection in sentence_divider.py.

PySBD is stable and works well, but its language support is limited to 23 languages. For a multilingual real-time voice interaction system, broader language coverage and more accurate sentence boundaries can improve TTS chunking and subtitle generation.

您期望的解决方案是什么？ / Describe the solution you'd like

Consider using yasbd-lib as the sentence boundary detector.

Example:

from yasbd import BoundaryDetector, get_supported_langs

SUPPORTED_LANGUAGES = get_supported_langs()

segmenter = BoundaryDetector(lang="en")
sentences = list(segmenter.segment(text))

yasbd-lib provides:

• Support for 39 languages
• Improved sentence boundary detection accuracy
• Faster segmentation performance
• Streaming support
• Native spaCy integration

A PySBD-compatible adapter is also available if keeping the current API style is preferred.

此功能为何对 Open-LLM-VTuber 很重要？ / Why is this important for Open-LLM-VTuber?

Sentence segmentation directly affects real-time voice interaction, subtitle timing, and TTS chunk generation.

Improving the segmentation engine can provide better multilingual support and more reliable sentence splitting for users across different languages.

您考虑过哪些替代方案？ / Describe alternatives you've considered

• Continue using PySBD with the current implementation.
• Maintain the current sentence segmentation logic.

您是否愿意参与开发此功能？ / Would you like to work on this issue?

Yes

补充信息 / Additional context

Comparison: https://dev.to/speed_k_7e1b449706e59e433/yasbd-lib-vs-pysbd-two-philosophies-of-sentence-boundary-detection-i88 Benchmarks: https://github.com/speedyk-005/yasbd-lib/tree/main/benchmarks

I would be happy to open a PR implementing this change if the maintainers are interested.