概览
@goodandready/dsh-tts
README / ZH
插件文档
目录摘要
dsh.pub 核对固定版本的组合包契约、运行时事实与分发语义；完整 README 请查看源仓库。
在 GitHub 阅读完整 READMELIMITATIONS
已知限制
* **Shared-GPU contention.** The LLM and the local TTS engine compete for the same GPU. Sentence-level streaming improves *scheduling* — speech starts earlier — but it does not make the model or the TTS faster. * **It does not solve model TTFT.** Time-to-first-token is a property of the model and hardware; live streaming only removes the wait *after* the first sentence exists. * **Inline code can sound awkward.** With `skipCode: true` the sanitizer deletes inline code spans, so a sentence built out of them can be spoken with gaps. This is upstream behaviour on both the live and settled paths, but live mode speaks more sentences so it is heard more often. * **Performance varies** by local model, TTS engine and GPU configuration. The numbers quoted in the release notes come from one specific local setup. ---