v1.1.3

Volcengine can now be pinned to recognize a single foreign language — French, German, Japanese, Korean, Spanish, Russian, or Arabic. Picking a specific language automatically switches to the higher-accuracy "streaming input" mode (a protocol limitation means the text appears all at once after you let go, rather than streaming in as you speak); picking "Automatic" keeps the mixed Chinese/English plus dialect recognition, and the default experience is unchanged.

Note that Volcengine can only recognize one language per session. For mixed-language speech or use overseas, Soniox is still the smoother choice — bidirectional streaming, mixed-language support, low latency.

French connector words like donc, voilà, enfin, quoi, and genre used to get deleted as filler (voilà also happens to share a name with the product, which made the deletion doubly awkward) — they're now kept. Existing users will have these false positives automatically removed from the filler-word list on first launch of the new version, no manual cleanup needed.